Intelligent navigation method and system for jejunum catheterization based on real-time ultrasonic contrast

Abstract

The application provides a jejunum catheterization intelligent navigation method and system based on real-time ultrasonic contrast, relates to the technical field of biological medical engineering, and comprises the following steps: collecting multiple data sets; performing unified time base alignment on the multiple data sets, and performing structure encapsulation and normalization processing to obtain jejunum catheterization navigation acquisition data frames; performing harmonic selective extraction on the contrast frames in the jejunum catheterization navigation acquisition data frames to obtain a harmonic selective enhancement result set; generating a tube body probability graph based on the harmonic selective enhancement result set to obtain a tube body candidate data set; constructing a candidate tube body space-time graph to generate a consistent tube body trajectory set; performing head positioning judgment and reliability evaluation on the consistent tube body trajectory set to determine a jejunum catheterization head positioning result; generating navigation prompt information based on the jejunum catheterization head positioning result, and outputting a navigation interaction result; performing navigation comprehensive risk evaluation on the navigation interaction result, and triggering a closed-loop self-adaption and degradation strategy according to the evaluation result.

Description

Technical Field

[0001] This invention relates to the field of biomedical engineering technology, and in particular to an intelligent navigation method and system for jejunal catheter placement based on real-time ultrasound contrast imaging. Background Technology

[0002] With the increasing prevalence of bedside ultrasound in clinical diagnosis and treatment, ultrasound-guided jejunal tube placement has gradually become an important means of enteral nutritional support for critically ill patients due to its advantages such as being non-invasive, real-time, and repeatable. However, jejunal tube placement requires precise advancement of the catheter into the upper jejunum within the complex anatomical environment, demanding a high level of experience from the operator.

[0003] Currently, existing technologies have proposed methods for assisting jejunal tube placement based on contrast-enhanced ultrasound. By injecting a contrast agent, the strong echo signal generated by the contrast agent within the lumen enhances the visualization of the tube. The flow of the contrast agent within the lumen helps determine the position of the tube tip. This contrast enhancement improves the contrast between the tube and surrounding tissues, thus improving tube visibility to some extent.

[0004] However, existing technologies rely on the operator's subjective interpretation and lack selective utilization of nonlinear harmonic evidence from contrast imaging. Under dynamic interference such as respiration, peristalsis, and gas obstruction, the enhanced signal of the tube is easily confused with artifacts such as blood vessels, intestinal walls, and gas, leading to a decrease in the accuracy of tube identification. At the same time, existing technologies are based on single-frame image judgment and lack cross-frame evidence aggregation and topological connectivity constraints. When the tube is obstructed or the imaging is uneven, trajectory chain breaks are easily generated, making it impossible to achieve stable tube tracking and continuous positioning of the head end. Summary of the Invention

[0005] To address the shortcomings of existing technologies, which rely on operator subjective interpretation and lack selective utilization of contrast-enhanced nonlinear harmonic evidence, and where enhanced tube signals are easily confused with artifacts such as blood vessels, intestinal walls, and gas under dynamic interferences like respiration, peristalsis, and gas obstruction, leading to decreased tube identification accuracy; and to further address the issues that existing technologies rely solely on single-frame image judgment, lacking cross-frame evidence aggregation and topological connectivity constraints, which can easily cause trajectory chain breaks when tubes are obstructed or imaging is uneven, making it impossible to achieve stable tube tracking and continuous tip positioning, this invention provides an intelligent navigation method and system for jejunal tube placement based on real-time contrast-enhanced ultrasound.

[0006] The technical solutions provided by the embodiments of the present invention are as follows: The first aspect of this invention provides a smart navigation method for jejunal catheter placement based on real-time ultrasound contrast imaging, comprising: S1: Collect multiple datasets; S2: Perform unified time base alignment on multiple datasets, and encapsulate and normalize the structure to obtain jejunal catheterization navigation acquisition data frames; S3: Perform harmonic selective extraction on the contrast frames in the jejunal catheterization navigation data frames to obtain the harmonic selective enhancement result set; S4: Based on the harmonic selective enhancement result set, generate the tube probability map and obtain the tube candidate dataset; S5: Based on the candidate tube dataset, construct the spatiotemporal graph of the candidate tubes and generate a consistent tube trajectory set; S6: Perform head-end positioning decision and reliability assessment on the consistent tube trajectory set to determine the jejunal tube head-end positioning result; S7: Based on the positioning results of the jejunal tube tip, generate navigation prompts and output navigation interaction results; S8: Perform a comprehensive navigation risk assessment on the navigation interaction results, and trigger a closed-loop self-adaptation and degradation strategy based on the assessment results.

[0007] A second aspect of the present invention provides an intelligent navigation system for jejunal catheter placement based on real-time ultrasound contrast imaging, comprising: processor; A memory storing computer-readable instructions, which, when executed by the processor, implement the intelligent navigation method for jejunal catheter placement based on real-time ultrasound contrast imaging as described in the first aspect.

[0008] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the intelligent navigation method for jejunal catheter placement based on real-time ultrasound contrast imaging as described in the first aspect.

[0009] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: In this invention, by performing harmonic selective extraction on the contrast frames, interference from the linear fundamental wave of tissue and the broadband strong echo of gas reflection is effectively suppressed, highlighting the nonlinear response of microbubbles and significantly enhancing the separability of tubes from artifacts such as blood vessels, intestinal walls, and gas. This overcomes the shortcomings of relying on subjective interpretation by the operator and the easy confusion of tube identification in complex sonographic environments. At the same time, by constructing a spatiotemporal map of candidate tubes and performing cross-frame evidence aggregation and topological connectivity constraints, the trajectory continuity can still be maintained even when the tube is obscured or the imaging is uneven. This overcomes the insufficiency of trajectory chain breakage caused by single-frame judgment, and achieves stable tube tracking and continuous head-end positioning, meeting the clinical needs for high-precision and high-reliability intelligent navigation. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating an intelligent navigation method for jejunal catheter placement based on real-time ultrasound contrast imaging, provided as an embodiment of the present invention.

[0012] Figure 2 This is a schematic diagram illustrating the temporal variation of head-end positioning coordinates and reliability, provided as an embodiment of the present invention.

[0013] Figure 3 This is a schematic diagram of a jejunal catheter placement intelligent navigation system based on real-time ultrasound contrast imaging, provided as an embodiment of the present invention. Detailed Implementation

[0014] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0015] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0016] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0017] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] Reference manual attached Figure 1 The diagram shows a flowchart of an intelligent navigation method for jejunal catheter placement based on real-time ultrasound contrast imaging, provided by an embodiment of the present invention.

[0020] This invention provides an intelligent navigation method for jejunal catheter placement based on real-time contrast-enhanced ultrasound. This method can be implemented using an intelligent navigation device for jejunal catheter placement based on real-time contrast-enhanced ultrasound, which can be a terminal or a server. The processing flow of the intelligent navigation method for jejunal catheter placement based on real-time contrast-enhanced ultrasound may include the following steps: S1: Collect multiple datasets.

[0021] Optionally, the various datasets specifically include: real-time B-mode ultrasound image dataset, real-time ultrasound contrast image dataset, harmonic and nonlinear enhancement feature dataset, imaging state parameter dataset, contrast agent preparation and injection event dataset, catheter placement operation context dataset, probe scanning window position and attitude dataset, and channel identification dataset.

[0022] Specifically, the real-time B-mode ultrasound image dataset includes: a sequence of grayscale images of tissue structures continuously output under the scanning window, frame numbers, and resolution information, which are used to provide background for anatomical structure reference and anchor point identification.

[0023] Furthermore, the real-time ultrasound contrast imaging dataset includes: enhanced image frame sequences, enhanced energy maps, or equivalent contrast display channel data output in contrast mode, used to obtain the enhancement differences caused by the contrast agent and improve the separability of the tube body from the background.

[0024] Furthermore, the harmonic and nonlinear enhancement feature dataset includes a set of feature quantities derived from the device side or calculated online by the acquisition end, such as harmonic energy, fundamental-harmonic ratio, nonlinear echo intensity, local enhancement gradient, and enhancement persistence. These serve as a calculable basis for selectively suppressing the linear fundamental wave of tissues and highlighting the nonlinear response of microbubbles at the source.

[0025] Furthermore, the imaging state parameter dataset includes state parameters that are bound to each frame of image, such as mechanical index (MI), total gain, dynamic range, TGC segmentation parameters, imaging depth, focal position, probe frequency range, angiography mode flag, and preset number, to support subsequent threshold self-adaptation and far-field penetration consistency constraints.

[0026] It should be noted that those skilled in the art can set the size of the preset number according to actual needs, and this invention does not limit this.

[0027] Among them, the mechanical index MI characterizes the sound pressure amplitude generated by the ultrasound beam in the tissue and is used as a safety indicator to assess the biological effects of ultrasound.

[0028] Among them, the TGC segmentation parameter refers to the time gain compensation segmentation parameter, which is used to adjust the segmented gain of echo signals at different depths to compensate for the attenuation of ultrasound waves when they propagate in tissues, so as to make the overall brightness of the image uniform.

[0029] Furthermore, the contrast agent preparation and injection event dataset includes: mixing batch number, mixing ratio setting, injection start and end events, injection volume, injection rate, flushing event, and contrast stage state machine markers driven by injection events, used to explicitly divide the contrast process into pre-filling, rising, stabilizing, and decaying stages and to define a stable imaging window for subsequent decisions.

[0030] Furthermore, the tube placement operation context dataset includes: tube advancement depth scale values, advancement, retraction, rotation and pause action codes, key action confirmation points and action durations, which are used to synchronously bind image evidence with the operation process.

[0031] Furthermore, the probe scanning window position and attitude dataset includes: window position labels for the antral window, pyloric window, duodenal window, and jejunal window, probe azimuth angle, pitch angle, roll angle attitude description, and scanning trajectory segment number, which are used to constrain the two-dimensional section to a reproducible scanning context.

[0032] Furthermore, the channel identifier dataset includes: device local timestamp, gateway receive timestamp, enqueue encapsulation timestamp, channel number, frame type marker, and frame sequence number, used for unified time base sorting and multi-source record alignment.

[0033] S2: Perform unified time base alignment on multiple datasets, and encapsulate and normalize the structure to obtain jejunal catheterization navigation acquisition data frames.

[0034] In one possible implementation, S2 specifically includes sub-steps S201 to S206: S201: Perform classification labeling and priority management on multi-class datasets, and perform time alignment of multi-class datasets through an edge acquisition gateway and a unified time base.

[0035] S202: Construct data frames from the aligned multi-class datasets to generate jejunal catheterization navigation acquisition data frames.

[0036] S203: Through the edge acquisition gateway, maintain the global acquisition queue and local queues, and perform queue organization and buffering of jejunal catheterization navigation acquisition data frames. Among them, the global acquisition queue is sorted according to a unified sorting time, and the local queue is divided according to channel and frame type.

[0037] S204: Perform linear scaling and normalization processing on the image data in the jejunal tube placement navigation acquisition data frame.

[0038] S205: Based on the processing results, dynamic range compression and background baseline correction are performed on the intensity of the contrast image to form a standardized enhancement intensity metric consistent with harmonic selective extraction.

[0039] S206: Based on standardized enhanced strength metrics, thresholding and hierarchical encoding are applied to the quality marker field to mark each data frame as either available or unavailable.

[0040] Specifically, the acquired multi-class datasets are classified, labeled, and prioritized. Frames marked as stable segments in the contrast-enhanced state machine are defined as stable imaging frames. Frames containing key window labels for the pyloric and duodenal regions are defined as key window frames. Stable imaging frames and key window frames are given high priority in the buffer queue to ensure they are preferentially entered into the computational path of harmonic selective extraction + spatiotemporal connectivity constraints. Unstable segments are marked as transition frames, used only for background modeling and motion compensation, and do not directly trigger head-end decisions.

[0041] Furthermore, the edge acquisition gateway uses a unified time base to perform time alignment on multiple datasets to obtain processed multi-class datasets. For each frame image and each event record, a local device timestamp, a gateway receiving timestamp, and an encapsulation queuing timestamp are simultaneously written. Delay compensation and out-of-order detection are performed based on link round-trip delay statistics and jitter statistics, remapping multi-source records to the same continuous time axis to obtain a unified sorting time. When a frame sequence number jump, timestamp rollback, or abnormal interval between adjacent frames is detected, a link quality field is written, triggering a frame loss filling flag or a session degradation flag, preventing the incorrect removal of real pipeline segments due to spatiotemporal connectivity constraints under incorrect timing.

[0042] Specifically, the data frames acquired during jejunal catheterization navigation include: session ID, frame number, frame type label, unified sorting time, raw B-mode image data or raw angiography image data, resolution and frame rate, harmonic and nonlinear enhancement feature vectors, imaging status parameter set, angiography batch number and scale, injection stage status, catheterization depth scale and action code, scan window label and probe attitude, link delay and frame loss count, and a quality marker field. The quality marker field is used to record far-field penetration score, gas obstruction ratio, motion intensity score, and enhancement stability score, so as to select valid frames to enter the catheter body probability map according to the quality threshold.

[0043] Furthermore, the jejunal catheterization navigation data frames are queued and cached: a global acquisition queue ordered by uniform time and a local queue divided by channel and frame type are maintained at the edge acquisition gateway. A priority retention strategy is set for stable imaging frames and key window frames, while a discard strategy is set for transitional frames. A circular buffer is used to limit the buffer level and record the queue level change curve. When the buffer approaches its upper limit, transitional frames are discarded first, while continuous stable imaging frames are retained, ensuring consistent inter-frame input without interruption in the thin, continuous line movement. The jejunal catheterization navigation data frames are linearly scaled and normalized, mapping imaging parameters such as MI, gain, dynamic range, depth, and focus to a unified range, eliminating dimensional differences caused by different device apertures. Dynamic range compression and background baseline correction are performed on the contrast image intensity to form a standardized enhancement intensity metric consistent with harmonic selective extraction. Injection rate, injection volume, and stage duration are standardized, making the contrast stage state machine comparable on a uniform scale. The insertion depth scale and action duration are standardized to form a computable joint observation vector of image frames, injection events, and operation actions under the same time index. The quality label field is thresholded and hierarchically encoded, marked as available or unavailable, for robust recognition and confidence-driven degradation of navigation output.

[0044] In this embodiment of the invention, by performing unified time base alignment and normalization on multiple datasets, the time deviation and dimensional differences of multi-source data are eliminated, providing jejunal catheterization navigation acquisition data frames that are time-consistent, standardized and comparable for subsequent steps.

[0045] S3: Perform harmonic selective extraction on the contrast frames in the jejunal catheterization navigation data frames to obtain a harmonic selective enhancement result set.

[0046] In one possible implementation, S3 specifically includes sub-steps S301 to S306: S301: Perform mode splitting, windowing reconstruction, and inter-frame registration on the data frame sequence of jejunal catheterization navigation acquisition data frames to obtain the contrast processing input set. The contrast processing input set includes a subset of contrast frames, a subset of B-mode reference frames, a subset of harmonic and nonlinear enhancement features, a subset of imaging state parameters, a subset of the contrast stage state machine, and a subset of quality markers.

[0047] S302: Using a subset of B-mode reference frames as the structural reference, the inter-frame displacement field is estimated through phase correlation.

[0048] S303: Apply the inter-frame displacement field to the subset of imaging frames and the subset of harmonic features to complete the pose registration and obtain the enhanced sequence with motion interference removed.

[0049] S304: Based on the enhancement sequence, an initial harmonic selective enhancement result set is generated by selectively extracting harmonics and suppressing linear fundamental waves.

[0050] S305: Based on the initial harmonic selective enhancement result set, construct the tissue echo baseline using the B-mode structure channel in the B-mode reference frame subset, and perform baseline subtraction on the contrast enhancement channel to obtain the enhancement residual map.

[0051] S306: Apply edge-preserving filtering and slender structure enhancement operators to the enhanced residual map, and output a set of harmonic selective enhancement results.

[0052] Specifically, the jejunal catheterization navigation acquisition data frame sequence is input, and the jejunal catheterization navigation acquisition data frame sequence is subjected to mode splitting, windowing reconstruction and inter-frame registration to obtain the contrast processing input set. Mode splitting is performed based on frame type label, contrast mode label and contrast stage state machine label. The mixed frame sequence within the session is split into contrast frame and harmonic frame subsequences, B-mode structure reference subsequence and transition and background subsequences according to imaging mode and purpose. The imaging state parameters, harmonic and nonlinear enhancement features, window label, probe posture and quality label bound to each frame are established one-to-one mapping according to a unified sorting time to avoid mixing grayscale aperture and contrast aperture of different modes.

[0053] Furthermore, windowed reassembly refers to segmenting and reassembling the angiography frame and harmonic frame subsequences using a unified sorting time as an index, according to a sliding time window or a fixed time window, to form an angiography window frame group containing sufficient temporal context. B-frames that overlap with or are closest to this window's time span are selected from the B-mode structural reference subsequence to form a structural reference frame group. Simultaneously, the sub-windows are pruned according to the angiography stage state machine, retaining only frames that meet the entry conditions to enter the main link of the window. Transitional frames are retained as auxiliary frames for motion estimation or baseline estimation but do not trigger a decision. Inter-frame registration refers to estimating the inter-frame displacement field within each angiography window using the structural reference frame or a designated reference frame as the coordinate reference, employing block matching, dense optical flow, or phase correlation. This displacement field is then synchronously applied to the angiography frame, harmonic correlation feature channel, and enhancement channel to complete in-position pose resampling, aligning the same tissue point to the same pixel coordinate system in different frames. Frames with excessively large registration residuals or motion intensity scores exceeding the upper limit are marked with high motion and have their weight reduced or temporarily excluded from the main computation path to suppress motion pseudo-enhancement strips from entering candidate generation.

[0054] Furthermore, the contrast processing input set includes: a subset of contrast frames, a subset of B-mode reference frames, a subset of harmonic and nonlinear enhancement features, a subset of imaging state parameters, a subset of the contrast stage state machine, and a subset of quality labels. The contrast frame subset includes: stable imaging frames labeled as contrast frames and harmonic frames and labeled as stable segments in the contrast stage state machine, as well as key window frames labeled as pyloric window, duodenal window, etc., used to enter the main computational path for harmonic selective extraction.

[0055] Specifically, the B-mode reference frame subset includes a sequence of B frames adjacent to the unified sequencing time of the contrast images, used to provide baseline and boundary references for tissue structure. The harmonic and nonlinear enhancement feature subset includes eigenvectors such as harmonic energy, fundamental-harmonic ratio, nonlinear echo intensity, local enhancement gradient, and enhancement persistence, used to form a calculable basis for suppressing linear components of tissue and highlighting the nonlinear response of microbubbles.

[0056] Furthermore, the imaging state parameter subset includes conditional parameters bound to each frame, such as mechanical index (MI), gain, dynamic range, depth, focus, and frequency range, used for conditional self-adaptation of thresholds and weights. The contrast-enhancing stage state machine subset includes markers for pre-filling, rising phase, stabilizing phase, and attenuation phase, used to limit extraction to triggering only within the stable imaging window. The quality marker subset includes far-field penetration score, gas occlusion ratio, motion intensity score, and enhanced stability score, used to filter unusable frames and assign weights to usable frames.

[0057] For example, using a subset of B-mode reference frames as the structural baseline, phase correlation is used to estimate the inter-frame displacement field. This displacement field is then applied to the angiography frame subset and the harmonic feature subset to achieve pose registration, resulting in an enhanced sequence free from motion interference. When the motion intensity score exceeds the motion upper limit, the corresponding frame is marked as a high-motion frame and its weight is reduced or it is temporarily excluded from the main extraction path to prevent pseudo-enhanced strips caused by motion from entering subsequent candidate generation.

[0058] Furthermore, based on the motion-disrupted enhancement sequence, harmonic selective extraction and tissue linear fundamental suppression are performed on the contrast images to generate a harmonic selective enhancement result set. Harmonic energy and fundamental-harmonic ratio are jointly gated, with the gating threshold adaptively adjusted according to MI, depth, and gain to suppress tissue linear fundamental and far-field speckle amplification. Directional consistency constraints are introduced into the local enhancement gradient to limit random jumps in enhancement boundaries. A minimum duration frame threshold is introduced into the enhancement persistence to eliminate short-term flicker enhancements. A tissue echo baseline is constructed using the B-mode structure channel, and baseline subtraction is performed on the contrast enhancement channel to obtain the enhancement residual map. Edge-preserving filtering and a slender structure enhancement operator are applied to the enhancement residual map to retain slender continuous responses and suppress blocky tissue enhancements. An occlusion mask is generated based on the gas occlusion ratio, and penalties or masking are applied to the bright stripes at the occlusion edges to prevent occlusion artifacts from entering the candidate list. The output harmonic selective enhancement result set includes: harmonic selective enhancement map, enhancement residual map, enhancement direction consistency map, enhancement persistence map, occlusion mask, unified sorting time, window label, imaging state condition vector and quality label, and is bound to the corresponding input frame for use in tube probability map generation and candidate quality score calculation.

[0059] It should be noted that those skilled in the art can set the size of the gate threshold according to actual needs, and this invention does not limit it.

[0060] In this embodiment of the invention, by performing harmonic selective extraction on the imaging frame, interference from tissue linear fundamental wave and gas reflection artifacts is effectively suppressed, highlighting the nonlinear response of microbubbles, significantly enhancing the separability of tubes from the background, and providing high-quality enhanced evidence for tube identification.

[0061] S4: Based on the harmonic selective enhancement result set, generate the tube probability map and obtain the tube candidate dataset.

[0062] In one possible implementation, S4 specifically includes sub-steps S401 to S407: S401: Based on the inter-frame displacement field in the harmonic selective enhancement result set, the harmonic selective enhancement map, enhancement residual map, enhancement direction consistency map, enhancement persistence map, and B-mode reference structure channel are resampled to the same pixel coordinate system to obtain the registered multi-channel feature set.

[0063] S402: Perform scale normalization and dynamic range compression on each channel of the multi-channel feature set, construct a condition vector formed by imaging state parameters, and conditionally weight and fuse each channel through the condition vector to form a fused feature tensor.

[0064] S403: Perform channel mixing on the fused feature tensor and extract local spatial features. Output the local spatial features to the segmentation head or detection head to obtain the tube body probability map.

[0065] S404: Based on the tube probability graph, extract the candidate connected component set and the candidate centerline seed set to generate the initial tube candidate dataset.

[0066] S405: Calculate the quality score for each candidate connected component in the initial tube candidate dataset. The specific calculation method for the quality score is as follows: The average tube confidence score is obtained by summing the tube probability values ​​of each pixel in the candidate region of the initial tube candidate dataset and dividing the sum by the total number of pixels in the candidate region.

[0067] The slenderness index of the candidate region is multiplied by the slenderness reward coefficient, and then an exponential calculation is performed to obtain the slenderness reward item.

[0068] The curvature penalty term is obtained by multiplying the curvature change rate index of the centerline of the candidate region by the curvature penalty coefficient, taking the negative value, and then performing an exponential operation.

[0069] Multiply the occlusion coverage ratio index by the occlusion penalty coefficient, take the negative value, and then perform an exponential operation to obtain the occlusion penalty term.

[0070] The quality score is obtained by multiplying the average tube confidence, elongation reward, curvature penalty, and occlusion penalty together.

[0071] S406: Based on the quality score and a quality threshold, the initial candidate dataset for pipe bodies is categorized and filtered. The categorized filtering specifically involves: Candidates with quality scores below the quality threshold are marked as first-level candidates and eliminated or downweighted. Candidates with quality scores not lower than the quality threshold are considered valid candidates and sorted from highest to lowest quality score. The top preset number of candidates are selected as main candidates.

[0072] It should be noted that those skilled in the art can set the size of the quality threshold according to actual needs, and this invention does not limit it.

[0073] S407: Perform morphological and physical prior constraint filtering on the filtered candidate tube connected component set and candidate centerline seed set, and output the tube candidate dataset.

[0074] Specifically, the input harmonic selective enhancement result set is used to resample the harmonic selective enhancement map, enhancement residual map, enhancement direction consistency map, enhancement persistence map and B-mode reference structure channel of the inter-frame displacement field in the harmonic selective enhancement result set to the same pixel coordinate system. Scale normalization and dynamic range compression processing are performed on each channel. A conditional vector formed by imaging state parameters is constructed and conditionally weighted and fused. Weights are generated for each channel. The weights are obtained from the conditional vector through linear mapping or multilayer perceptron and normalized by Softmax, so that the contribution of each channel is self-adaptive under different MI, different gains and different depth conditions.

[0075] Softmax normalization is a function that transforms multiple numerical values ​​into a probability distribution. It works by performing an exponential operation on each numerical value and then dividing it by the sum of the exponents of all the numerical values, so that the output result is between 0 and 1 and the sum is 1.

[0076] Further, each channel is weighted according to its weight to obtain a weighted channel. The weighted harmonic selectivity enhancement map, enhancement residual map, orientation consistency map, and persistence map are then concatenated with the B-mode structure channel in the channel dimension to form a fused feature tensor. Convolution is performed on the fused feature tensor to perform channel mixing and extract local spatial features. The output is then fed into the segmentation head or detection head to obtain the tube probability map, generating a tube candidate dataset. The tube candidate dataset includes: the tube probability map, a set of candidate tube connected regions, a seed set of candidate centerlines, and a set of candidate quality scores. The tube probability map includes the probability value of each pixel belonging to a tube, obtained by concatenating the harmonic selectivity enhancement map, enhancement residual map, orientation consistency map, persistence map, and B-mode structure channel and inputting it into the segmentation network or detection network. The imaging state condition vector is injected into the network using conditional normalization or gated weights to ensure that the output probability is within a comparable range under different MI / gain / depth conditions. The candidate tube connected region set includes the set of candidate regions obtained by performing adaptive threshold segmentation and connected region extraction on the tube probability map. The candidate centerline seed set includes: a set of initial centerline points, endpoints, and principal direction vectors generated by skeletonization, minimum path, or refined tracing of the candidate regions. These are used to provide initial trajectories for the spatiotemporal connectivity constraints of subsequent continuous strip-shaped traversal. The candidate quality score set includes: a quality score value calculated for each candidate connected component, which serves as the basis for candidate ranking and gating.

[0077] Furthermore, based on the quality score, a hierarchical filtering and weighting process is performed on the candidate connected component set in the candidate tube dataset. The probability values ​​of each pixel belonging to a tube in the region are summed and then divided by the total number of pixels contained in the candidate region to obtain the average tube confidence score. The elongation index of the candidate region is multiplied by the elongation reward coefficient and then exponentially calculated to obtain the elongation reward term. The curvature change rate index of the candidate region's centerline is multiplied by the curvature penalty coefficient, negatively taken, and then exponentially calculated to obtain the curvature penalty term. The occlusion coverage ratio index is multiplied by the occlusion penalty coefficient, negatively taken, and then exponentially calculated to obtain the occlusion penalty term. The quality score of the candidate region is calculated by multiplying the average tube confidence score, elongation reward term, curvature penalty term, and occlusion penalty term together.

[0078] Specifically, the formula for calculating the quality score is as follows: in, Let represent the quality score of the connected component of the k-th candidate tube, used for stable sorting, gating, and weighted fusion of candidate tubes. The number of pixels in the connected component of the k-th candidate tube is obtained by adaptive thresholding of the tube probability map and performing connected component analysis. It is used to calculate the elongation, curvature and occlusion ratio regional features. The probability value representing the pixel (x, y) belonging to the tube is obtained by inputting channel features into the segmentation network and detection network and outputting via Sigmoid. The value range is [0, 1], and it is used to characterize the tube consistency strength within the candidate region. The slenderness reward coefficient is determined by parameter sweeping and cross-validation of candidate ranking consistency, false positive rate, and false negative rate. It is a real number greater than 0 and is used to adjust the improvement of the slenderness index on the quality score. The slenderness index of the k-th candidate is calculated by the geometric principal axis and secondary axis through the candidate connected components and obtained by the ratio of the centerline length to the equivalent width. It is used to characterize whether the candidate conforms to the morphological prior of the tube's slender strip-like course. The curvature penalty coefficient is determined by fitting and sweeping the curvature distributions of real tube candidates and artifact candidates. It is a real number greater than 0 and is used to adjust the suppression strength of the curvature change rate on the score. The curvature change rate index of the k-th candidate centerline is obtained by calculating the direction angle of adjacent line segments for the candidate centerline seed and taking the norm of the angle increment. It is used to quantify whether the centerline is smooth and continuous. This represents the occlusion penalty coefficient, which is determined by scanning the false detection sensitivity caused by gas occlusion and strong reflection edge stripes. It is a real number with a value greater than 0 and is used to control the rate at which the occlusion ratio decays the score. The occlusion coverage ratio index represents the k-th candidate, which is obtained by calculating the proportion of occluded pixels in the candidate area on the occlusion mask. It is used to penalize areas where false candidates are easily generated, such as the bright stripes at the occlusion edge and the gas reflection boundary.

[0079] Furthermore, when the quality score is lower than the quality threshold, the corresponding candidate is marked as a first-level candidate and eliminated or downweighted. When the quality score is not lower than the quality threshold, the corresponding candidate is considered a valid candidate, and the top few candidates are selected as main candidates based on their quality scores from high to low. Morphological and physical prior constraints are applied to the candidate tube connected component set and candidate centerline seed set in the tube candidate dataset. Based on the candidate quality score set, the candidates are stably sorted, and candidates with quality scores not lower than the threshold are preferentially retained to enter the spatiotemporal connectivity constraint path. Candidates with elongation indices or centerline lengths below the lower limit are marked as strong echo candidates and eliminated. Candidates with curvature change rates higher than the upper limit or multiple bifurcations are marked as non-tube reflective stripes and their weights are reduced or eliminated. Candidates with occlusion coverage ratios higher than the occlusion upper limit are subject to occlusion penalties or directly eliminated to prevent occlusion edge strips from mistakenly entering the candidate pool. Output the tube candidate dataset, which is output frame by frame according to a uniform sorting time and bound to the window label, imaging state condition vector, angiography stage marker and quality marker fields.

[0080] In this embodiment of the invention, by generating a tube probability map and filtering to obtain a tube candidate dataset, the enhanced evidence is transformed into a quantifiable tube confidence score, and isolated strong echoes and short artifacts are eliminated, providing highly reliable candidate inputs for subsequent spatiotemporal tracking.

[0081] S5: Based on the candidate tube dataset, construct the spatiotemporal graph of the candidate tubes and generate a consistent tube trajectory set.

[0082] In one possible implementation, S5 specifically includes sub-steps S501 to S505: S501: Based on the candidate dataset of the tube body, using the unified sorting time as the index, extract the candidate set of the previous frame and the candidate set of the next frame from the candidate dataset of the tube body, and establish candidate index tables respectively.

[0083] S502: Based on the candidate index table, construct a cross-frame cost matrix for each pair of candidates in adjacent frames. Each element in the cross-frame cost matrix represents the association cost between the candidate in the previous frame and the candidate in the next frame.

[0084] S503: Based on the cross-frame cost matrix, a one-to-one matching is performed on candidates using a minimum cost matching strategy. Specifically, the matching rules are as follows: a candidate at the same time step is only allowed to match one candidate at the next time step, and a candidate at the same time step is only allowed to be matched by one candidate at the previous time step.

[0085] S504: Write the successfully matched candidate pairs as directed edges into the edge set, and write edge attributes such as cost, prediction error, direction consistency, shape similarity and occlusion penalty for each edge to form a candidate tube spatiotemporal graph.

[0086] S505: Based on the spatiotemporal graph of the candidate tube, spatiotemporal connectivity constraints are applied to the candidate tubes to generate a consistent tube trajectory set.

[0087] Specifically, the candidate tube spatiotemporal graph includes a directed graph structure with candidate centerline seeds or candidate connected components as nodes and candidate associations between adjacent frames as edges. Node attributes include at least: candidate region number, candidate centerline point sequence, endpoint coordinates, principal direction vector, candidate region quality score, average tube confidence score of candidate region, elongation index, rate of curvature change index, and occlusion coverage ratio index.

[0088] Specifically, the edge attributes include at least: endpoint displacement distance, centerline shape similarity, principal direction angle difference, candidate region overlap rate, and consistency difference between tube probability map similarity and quality score. Using a unified sorting time as an index, candidate index tables are established for the previous and next frame candidate sets respectively, and a cross-frame cost matrix is ​​extracted for each candidate. Position prediction is performed on the previous frame candidates based on the inter-frame displacement field, remapping the predicted endpoint coordinates and predicted centerline point sequence to the next frame coordinate system. A gating radius or gating box is set around the predicted position to limit the candidate pair search space. The gating range is adaptively adjusted according to imaging depth, frame rate, and motion intensity score; candidate pairs exceeding the gating range are determined to be unmatched. The cross-frame association cost is calculated for candidate pairs within the gating range. The cross-frame association cost is obtained by weighted summation of endpoint displacement difference, main direction angle difference, centerline shape difference, candidate region overlap difference, tube probability consistency difference and quality score consistency difference. An additional occlusion penalty is added for candidate pairs with an occlusion coverage ratio higher than the occlusion limit, and an additional artifact penalty is added for candidate pairs with a curvature change rate higher than the curvature limit.

[0089] Furthermore, any element in the cross-frame cost matrix represents the association cost between "Candidate A in the previous frame" and "Candidate B in the next frame." The association cost function can be expressed as follows: the cross-frame association cost equals the normalized amount of the endpoint displacement difference multiplied by the endpoint displacement weight, plus the normalized amount of the main direction angle difference multiplied by the direction weight, plus the normalized amount of the centerline shape difference multiplied by the shape weight, plus the normalized amount of the candidate region overlap difference multiplied by the overlap weight, plus the normalized amount of the tube probability consistency difference multiplied by the probability weight, plus the normalized amount of the quality score consistency difference multiplied by the score weight. When the candidate occlusion coverage ratio exceeds the occlusion limit, an occlusion penalty term is added; when the candidate curvature change rate exceeds the curvature limit, an artifact penalty term is added. The normalized amount of each difference term is obtained by linearly normalizing the minimum and maximum values ​​of the corresponding indicators within the candidate pair gate range, and the penalty term is obtained by monotonically increasing the mapping of the over-limit amplitude. One-to-one matching is performed based on the cross-frame cost matrix to generate a set of associated edges. For a candidate at the same time, it is only allowed to match one candidate at the next time, and the candidate at the next time is only allowed to be matched by one candidate at the previous time. The minimum cost matching strategy is used to achieve one-to-one association.

[0090] Furthermore, when the minimum association cost is higher than the cost threshold, it is determined that the candidate does not have reliable continuation in the next frame, a trajectory breakpoint is generated, and a bridging window is initiated. When a candidate has no match for several consecutive frames and its quality score is lower than the quality threshold, the candidate is marked as an isolated artifact node and removed from the spatiotemporal graph. Thus, successfully matched candidate pairs are written into the edge set as directed edges, and edge attributes such as cost, prediction error, direction consistency, shape similarity, and occlusion penalty are written to the edges, forming a candidate tube spatiotemporal graph. Based on the candidate tube spatiotemporal graph, spatiotemporal connectivity constraints are applied to the candidates to obtain a consistent tube trajectory set. The connectivity cost between candidates in adjacent frames is calculated, and a minimum total cost path search or dynamic programming is performed. The connectivity cost is obtained by weighted summation of endpoint displacement distance, main direction angle difference, centerline shape difference, candidate region overlap rate, tube probability graph similarity, and quality score difference. An occlusion penalty term is applied to candidates with a high occlusion coverage ratio. Within the sliding time window, candidate sequences with consistently high quality scores and continuous centerline movement are preferentially selected as the main trajectory. For time slices with dropped or high-motion frames, trajectory bridging is performed: within the bridging window, several missing candidate frames are allowed, and the endpoint positions predicted by extrapolation are matched and connected with the candidate endpoints of the next frame, and the bridging segment is marked as a first-level confidence interval. For detected isolated strong echo blocks, short-segment strafe reflections, and multi-bifurcation artifact candidate sequences, chain breaking is performed: when the candidate centerline length is continuously below the lower limit or the rate of curvature change is continuously above the upper limit, the sequence is determined to be an artifact sequence and removed from the spatiotemporal graph.

[0091] It should be noted that those skilled in the art can set the cost threshold according to actual needs, and this invention does not limit it.

[0092] Specifically, the consistent tube trajectory set includes: the main trajectory centerline sequence, the main trajectory endpoint sequence, the trajectory continuity score, the trajectory consistency score, and the trajectory reliability level label. The trajectory endpoint sequence is used for head-end positioning decision, and the trajectory consistency score is used to characterize the morphological consistency and endpoint motion consistency of the trajectory within the time window, and serves as one of the inputs to the head-end reliability assessment model.

[0093] In this embodiment of the invention, by constructing a spatiotemporal map of candidate tubes and performing spatiotemporal connectivity constraints, cross-frame evidence aggregation and topology preservation of continuous strip-shaped movement are achieved, maintaining trajectory continuity when tubes are occluded or unevenly developed, thus solving the problem of trajectory chain breakage caused by single-frame judgment.

[0094] S6: Perform head-end positioning decision and reliability assessment on the consistent tube trajectory set to determine the head-end positioning result of the jejunal tube placement.

[0095] In one possible implementation, S6 specifically includes: S601: Based on the consistent tube trajectory set and the harmonic selective enhancement result set, extract the local evidence vectors of the two endpoints of the main trajectory in each frame or each sliding time window.

[0096] S602: Based on the local evidence vector, calculate the endpoint cost value for each endpoint of the main trajectory. The specific calculation method for the endpoint cost value is as follows: Based on the mean probability of the tube body, the mean harmonic enhancement, the enhancement persistence score, the anchor point consistency score, and the candidate quality score in the local evidence vector, the first type of squared deviation term is obtained by subtracting the deviation from one and squaring them.

[0097] The occlusion coverage ratio and position jitter are squared to obtain the second type of squared deviation term.

[0098] The endpoint cost is obtained by weighted summation of the first type of squared deviation term and the second type of squared deviation term.

[0099] S603: By mapping the endpoint cost to the probability of the endpoint head using the temperature parameter, the head-end reliability assessment value is obtained. The specific calculation method for the head-end reliability assessment value is as follows: Divide the endpoint value of the first endpoint by the temperature parameter, take the negative value, and then perform an exponential operation to obtain the exponential weight of the first endpoint.

[0100] Divide the endpoint value of the second endpoint by the temperature parameter, take the negative value, and then perform an exponential operation to obtain the exponential weight of the second endpoint.

[0101] Add the index weights of the first and second endpoints to obtain the sum of the index weights of the endpoints.

[0102] Divide the endpoint value of any endpoint by the temperature parameter, take the negative value, and then perform an exponential operation to obtain the exponential weight of that endpoint.

[0103] Divide the exponential weight of this endpoint by the sum of the exponential weights to obtain the head-end credibility assessment value.

[0104] S604: The endpoint with the highest confidence evaluation value is determined as the head endpoint, and the coordinates of the endpoint are output as the head coordinates, and the tangential direction of the centerline at the endpoint is output as the head direction vector.

[0105] S605: In conjunction with the trust threshold, when the headend trust assessment value is lower than the trust threshold or the endpoint jitter is higher than the jitter upper limit threshold, the headend status is marked as restricted or unavailable, and a navigation degradation strategy is triggered.

[0106] It should be noted that those skilled in the art can set the size of the confidence threshold and the upper limit threshold according to actual needs, and this invention does not limit them.

[0107] S606: Outputs the jejunal tube placement tip localization result, including tip coordinates, tip orientation vector, tip confidence assessment value, and tip status marker.

[0108] In this embodiment of the invention, by performing credibility assessment and head-end positioning decision on the two endpoints of the trajectory, the probability of the endpoints being the head end is quantified and the state is dynamically marked. When the interference increases, degradation is automatically triggered to ensure the accuracy and reliability of the head-end positioning.

[0109] S7: Based on the positioning results of the jejunal tube tip, generate navigation prompts and output navigation interaction results.

[0110] In one possible implementation, S7 specifically includes sub-steps S701 to S705: S701: Based on the set of jejunal tube tip positioning results, the set of harmonic selective enhancement results, and the tube body probability map, the unified sorting time in the jejunal tube navigation acquisition data frame is used as the unique time index.

[0111] S702: Based on the unified sorting time, the head-end positioning results, trajectory endpoint sequences, window labels, harmonic selective enhancement maps, tube probability maps, occlusion masks, and quality marker fields under the same time index are aligned and jointly retrieved to obtain multimodal navigation data.

[0112] S703: Within each time slice corresponding to a unified sorting time, based on multimodal navigation data, read the head coordinates and head direction vector of the current time slice, and map the head coordinates and head direction vector to the pixel coordinate system of the current display frame.

[0113] S704: Combining the headend status flag, occlusion coverage ratio, and link quality field of the current time slice, when the headend status flag is restricted or unavailable, or the occlusion coverage ratio exceeds the occlusion limit, or the link quality field indicates the presence of frame loss filling flag and timestamp abnormal flag, the prompt will be downgraded to only outputting scan adjustment prompt and quality alarm prompt, and freezing the management operation prompt and headend direction guidance.

[0114] S705: Output navigation interaction results time-by-time, using a uniform sorting time as the time granularity. The navigation interaction results include an overlay guidance layer and a prompt level corresponding to each time slice.

[0115] Specifically, the endpoint local evidence vector includes the mean probability of the tube in the endpoint neighborhood, the mean harmonic selective enhancement intensity in the endpoint neighborhood, the enhancement persistence score in the endpoint neighborhood, the consistency score between the endpoint and the anatomical anchor point, the coverage ratio of the endpoint neighborhood occlusion mask, the positional jitter of the endpoint within the sliding time window, and the quality score of the candidate or trajectory segment to which the endpoint belongs. The endpoint neighborhood is a window centered on the endpoint coordinates, and the window size is adaptively set according to the imaging depth. The endpoint positional jitter is obtained by normalizing the mean square displacement or variance of the endpoint coordinates over several consecutive frames. The consistency score between the endpoint and the anatomical anchor point is obtained by mapping the distance, direction, and confidence interval matching between the endpoint and the preset anchor point region.

[0116] It should be noted that those skilled in the art can set the size of the preset anchor point area according to actual needs, and this invention does not limit this.

[0117] Furthermore, based on the local evidence vector at the endpoints, the mean probability of the pipe body, the mean harmonic enhancement, the enhancement persistence score, the anchor point consistency score, and the candidate quality score are each squared after subtracting the deviation from one. The occlusion coverage ratio and the positional jitter are directly squared, and the squared deviation terms after weighted summation are all added together to obtain the endpoint cost value. The endpoint costs at both ends are mapped to the probability that the endpoint is the head end, and the one with the highest probability is used as the head end credibility assessment value. The endpoint cost value of endpoint one is divided by the temperature parameter, the negative is taken, and then an exponential operation is performed to obtain the exponential weight of endpoint one. The endpoint cost value of endpoint two is divided by the temperature parameter, the negative is taken, and then an exponential operation is performed to obtain the exponential weight of endpoint two. Then the exponential weights of endpoint one and endpoint two are added to obtain the sum of the exponential weights of the two endpoints. The endpoint cost value of any endpoint is divided by the temperature parameter, the negative is taken, and then an exponential operation is performed to obtain the exponential weight of any endpoint. The exponential weight of any endpoint is divided by the sum of the exponential weights of the two endpoints to obtain the head end credibility assessment value.

[0118] Specifically, the formula for calculating the head-end credibility assessment value is as follows: in, This represents the head-end confidence assessment value of the i-th endpoint when the uniform sorting time is t, which is used for soft decision selection between the two endpoints. This represents the endpoint cost of the i-th endpoint when the unified sorting time is t. It is obtained by weighted penalty summation of the local evidence vectors of the endpoints. The value range is a real number greater than or equal to zero. It is used for sorting, gating and anomaly suppression of endpoints. The temperature parameter is determined by parameter sweeping and cross-validation of the head-end decision accuracy, misjudgment rate and positioning jitter. It is a real number with a value greater than zero and is used to adjust the sharpness of the mapping from the endpoint cost to the probability that the endpoint is the head end.

[0119] Furthermore, the endpoint that maximizes the head-end confidence assessment value is designated as the head-end endpoint, the output endpoint coordinates are used as the head-end coordinates, and the tangential direction of the centerline at the output endpoint is used as the head-end direction vector. When the head-end confidence assessment value is lower than the confidence threshold or the endpoint jitter exceeds the jitter upper limit threshold, the head-end status is marked as restricted or unavailable, triggering a navigation degradation strategy. The jejunal tube placement head-end positioning result is output. The jejunal tube placement head-end positioning result includes: unified sorting time, head-end coordinates, head-end direction vector, head-end confidence assessment value, head-end status marker, and positioning record bound to the window label, used for navigation prompts and interactive output.

[0120] For example, the window label, endpoint cost, final selected headend, reliability score, and status marker at different unified sorting times are used to quantify the operational reliability level of the headend device. In one unified sorting time, the headend reliability assessment value is 2025-09-01 at 10:00:03, with the window label being pyloric window. The endpoint cost of endpoint one is 1.2, and the endpoint cost of endpoint two is 2.8. The final selected headend is endpoint one, with a headend reliability assessment value of 0.65. The headend status marker is "limited (critical)," indicating that the reliability of the headend device is in a critically limited state at that moment. In another unified sorting time, the headend reliability assessment value is 2025-09-01 at 10:00:05, with the window label being duodenal window. The endpoint cost of endpoint one is 0.8. The endpoint cost of endpoint two is 3.2, the headend reliability assessment value. Endpoint one was ultimately selected. Its headend reliability assessment value is 0.72. The headend status is marked as available, indicating that the headend device's reliability meets normal operation requirements at this time. The unified sorting time is 2025-09-01 (headend reliability assessment value 10:00:07): window label is duodenal window. The endpoint cost of endpoint one is 0.6, the headend reliability assessment value is 3.5, and the final selected headend is endpoint one. Its headend reliability assessment value is 0.76. The headend status is marked as available, indicating that the headend device's reliability is at a good usable level at this time. The unified sorting time is 2025-09-01 (headend reliability assessment value 10:00:09): window label is jejunal window. The endpoint cost of endpoint one is 1.5, the headend reliability assessment value. The endpoint cost of endpoint two is 2.2, the headend is ultimately selected as endpoint one. The headend reliability assessment value is 0.57. The headend status is marked as restricted, indicating that the reliability of the headend device at this moment is lower than the normal operating standard. The unified sorting time is the headend reliability assessment value 2025-09-01, the headend reliability assessment value 10:00:11: the window label is jejunal window. The endpoint cost of endpoint one is 0.9, the headend reliability assessment value. The endpoint cost of endpoint two is 2.6, the headend reliability assessment value is 2.66, the headend is ultimately selected as endpoint one. The headend reliability assessment value is 0.66. The headend status is marked as restricted (critical), indicating that the reliability of the headend device at this moment is in a restricted state that is critical to meet the standard. The unified sorting time is based on the headend credibility assessment value of 2025-09-01 (headend credibility assessment value 10:00:13): the window label is jejunal window. The endpoint cost of endpoint one is 0.5 (headend credibility assessment value). The endpoint cost of endpoint two is 3.8 (headend credibility assessment value). Endpoint one is ultimately selected as the headend.The headend reliability assessment value is 0.79. The headend status is marked as available, indicating that the headend device's reliability is at a high availability level at this moment. The unified sorting time is 2025-09-01 (headend reliability assessment value 10:00:15): the window label is jejunal window. The endpoint cost of endpoint one is 1.1 (headend reliability assessment value). The endpoint cost of endpoint two is 2.9 (headend reliability assessment value). The final selected headend is endpoint one. The headend reliability assessment value is 0.67. The headend status is marked as restricted (critical), indicating that the headend device's reliability is in a restricted state that is critically available at this moment.

[0121] Reference manual attached Figure 2 The diagram illustrates the temporal variation of head-end positioning coordinates and reliability provided by the present invention.

[0122] Specifically, using the frame sequence range as the horizontal axis, the head-end localization results within each time slice are summarized and displayed at a uniform sorting time granularity: the blue curve represents the horizontal coordinate of the head-end, and the green curve represents the vertical coordinate of the head-end. It can be seen that both change continuously and slowly monotonically as the frame segment progresses, indicating that the head-end trajectory maintains good spatial coherence and directional consistency within the time window, without obvious jumps or reverse drift. The red curve represents the head-end confidence assessment value, which gradually decreases as the time slice progresses, and gradually approaches and falls below the usable threshold (dashed line) in the later stages. This reflects that as motion intensifies, the occlusion ratio increases, or the development quality fluctuates, the consistency between local evidence at the endpoint and trajectory evidence is weakened, causing the head-end decision to evolve from "usable" to "restricted / unusable." When the confidence level falls below the threshold, the corresponding time slice should trigger a downgrade prompt and re-capture guidance, freeze high-intensity tube placement operation prompts, and only retain scan adjustments and quality alarms to avoid misleading navigation output in the low-confidence range.

[0123] For example, using the unified sorting time in the jejunal catheterization navigation data frame as the unique time index, the head-end positioning results, trajectory endpoint sequences, window labels, harmonic selective enhancement maps, tube body probability maps, occlusion masks, and quality marker fields under the same time index are aligned and jointly retrieved. Within each time slice corresponding to the unified sorting time, the head-end coordinates and head-end direction vector of the time slice are first read and mapped to the pixel coordinate system of the current display frame to generate an overlay guidance layer. The overlay guidance layer includes at least head-end marker points, direction arrows, the main trajectory centerline trajectory, candidate region boundaries, and occlusion warning regions. Then, the head-end confidence assessment value, head-end status marker, motion intensity score, occlusion coverage ratio, far-field penetration score, and enhancement consistency score quality marker fields of the time slice are read to complete the warning classification, and the classification results are converted into text warning levels and audio-visual warning levels. Then, read the window position labels and window position switching records of the time slices, combine them with the displacement trend and direction consistency of the head end in several consecutive time slices with the same sorting, generate suggested scanning and adjustment prompts, and combine the tube insertion depth scale and action code to generate suggested tube insertion operation prompts.

[0124] Specifically, the occlusion limit is generated by a threshold rule for the occlusion coverage ratio: within each time slice, the occlusion coverage ratio is obtained by intersecting the occlusion mask with the candidate region or endpoint neighborhood window, and a sliding statistical baseline for the occlusion coverage ratio is maintained within the session. When the occlusion coverage ratio exceeds the threshold determined by the baseline mean plus the deviation tolerance, or exceeds the threshold for several consecutive time slices, it is determined that the occlusion limit has been exceeded and an occlusion exceedance flag is written. The occlusion limit is used to trigger a downgraded prompt and freeze the ducting operation prompt to avoid misleading navigation induced by the highlighted strips at the occlusion edge. When the header status of a time slice is marked as restricted or unavailable, or the occlusion coverage ratio exceeds the occlusion limit, or the link quality field indicates the presence of a frame loss filling flag and a timestamp anomaly flag, the prompt is automatically downgraded to only outputting a scan adjustment prompt and a quality alarm prompt, and the ducting operation prompt and header directional guidance are frozen to avoid misleading operations. Navigation interaction results are output slice by slice in a uniformly sorted time. This ensures that every prompt, every overlay display, and every audio-visual event is bound to a unique time index, guaranteeing that the prompts can be replayed and reproduced, are auditable and traceable, and are strictly aligned with the state machine of the angiography stage, window switching, and tube placement process.

[0125] Furthermore, the navigation prompts include at least: head-end coordinates, head-end direction vector, head-end confidence assessment value, head-end status marker, current window position label, suggested scan and adjustment prompts, and suggested placement operation prompts. Navigation interaction results include: frame-by-frame overlay guidance layers, text prompt sequences, audio-visual prompt event sequences, and prompt records bound to a unified sorting time. These prompt records are written to the session log and correspond one-to-one with the head-end positioning records for playback verification and quality traceability.

[0126] Furthermore, the link quality field includes at least a frame sequence number continuity flag, a timestamp monotonicity flag, an adjacent frame interval anomaly flag, a buffer overflow risk flag, and a frame loss count. The acquisition unit maintains the previous frame sequence number and the previous frame's unified sorting time for each channel. When a frame sequence number jump, duplication, or rollback is detected, or when a unified sorting time rollback, stagnation, or an interval abrupt change exceeding the allowable range is detected, the corresponding flag is set to anomaly and written to the link quality field. The link quality field is used to guide the degradation strategy for registration, candidate association, and navigation output.

[0127] Specifically, the rules for generating frame loss filling markers are as follows: When a gap in the frame sequence number is detected and the gap length does not exceed the allowable filling window, a placeholder frame is inserted on the unified sorting time axis, inheriting the structural reference information of the previous available frame, and the time slice is marked as a frame loss filling frame. The placeholder frame does not participate in the main decision calculation of the tube probability map; it is only used to maintain the spatiotemporal map connectivity and time axis continuity. When the gap length exceeds the allowable filling window, no filling is performed, but a trajectory breakpoint is directly generated and a bridging window is triggered. The rules for generating timestamp anomaly markers are as follows: The acquisition end performs consistency checks on the device's local timestamp, the gateway's received timestamp, and the encapsulated queuing timestamp. When any timestamp shows a rollback, a leap, or the relative difference between the three exceeds the allowable jitter band, the frame is marked as a timestamp anomaly. Timestamp anomaly frames are downweighted or removed during candidate cross-frame association and prompt output, triggering a conservative mode that only outputs quality alarms and scan adjustment prompts.

[0128] In this embodiment of the invention, by generating an overlay guidance layer and hierarchical navigation prompts, the head positioning results are presented to the operator in an intuitive way, and the prompts are automatically downgraded when the quality is insufficient, so as to avoid misleading operations and improve the safety of clinical use.

[0129] S8: Perform a comprehensive navigation risk assessment on the navigation interaction results, and trigger a closed-loop self-adaptation and degradation strategy based on the assessment results.

[0130] In one possible implementation, S8 specifically includes sub-steps S801 to S809: S801: Input the head-end credibility assessment value, head-end status marker, trajectory continuity score, and trajectory consistency score as input parameters for the comprehensive navigation risk assessment.

[0131] S802: Calculate the basic evidence strength and basic interference strength based on the input parameters.

[0132] S803: Multiply the basic evidence strength, the evidence contradiction penalty factor, and the conditional gating inhibition factor to obtain the evidence synthesis quantity.

[0133] S804: Input the evidence synthesis quantity into the evidence shaping operator to obtain the evidence-side output.

[0134] S805: Input the basic interference strength, interference coupling value and time fluctuation penalty value into the interference aggregation operator to obtain the interference side output.

[0135] S806: Add a very small positive real constant to the evidence-side output and the interference-side output respectively, and divide the processed interference-side output by the processed evidence-side output to obtain the interference term.

[0136] S807: After taking the natural logarithm of the interference term, input the monotonically bounded mapping function to obtain the comprehensive navigation risk assessment value.

[0137] S808: Compares the comprehensive navigation risk assessment value with different risk thresholds. When the comprehensive navigation risk assessment value is not higher than the secondary risk threshold, it is determined to be in normal navigation mode, and outputs a management operation prompt and a scan prompt. When the comprehensive navigation risk assessment value is between the secondary risk threshold and the primary risk threshold, a degradation strategy is triggered. When the comprehensive navigation risk assessment value is higher than the primary risk threshold or the system is unavailable for several consecutive frames, a session protection strategy is triggered.

[0138] It should be noted that those skilled in the art can set the size of the primary risk threshold and the secondary risk threshold according to actual needs, and this invention does not limit them.

[0139] S809: Output adaptive control results. These results include the navigation mode flag and the corresponding strategy control parameters.

[0140] Specifically, the self-adaptation and degradation strategies include at least: frame-level gating strategy, prompt degradation strategy, recapture strategy, and session protection strategy. The frame-level gating strategy determines which frames enter the prompt generation link; the prompt degradation strategy determines whether the prompt is downgraded from a managed operation prompt to a scan-only prompt; the recapture strategy guides the reacquisition of available window positions and imaging windows when the headend is lost; and the session protection strategy freezes the headend output under low-quality conditions to avoid erroneous guidance. The evidence synthesis quantity is obtained by multiplying the basic evidence strength by the evidence contradiction penalty factor and the conditional gating suppression factor. The evidence synthesis quantity is input into the evidence shaping operator to obtain the evidence-side output. The basic interference strength, interference coupling value, and time fluctuation penalty value are input together into the interference aggregation operator to obtain the interference-side output. A very small positive real constant is added to both the evidence-side output and the interference-side output. The interference-side output is divided by the evidence-side output to obtain the interference term. The interference term is taken as its natural logarithm and then input into a monotone bounded mapping function to obtain the navigation comprehensive risk assessment value. The specific calculation formula for the navigation comprehensive risk assessment value is as follows: in, This represents the comprehensive navigation risk assessment value at time t, which is used to quantitatively assess the reliability of the current navigation output and trigger closed-loop gating, degradation prompts, recapture, and session protection. The strength of basic evidence is obtained by normalizing the observable quality item obtained by fusing the head-end credibility assessment value, trajectory continuity score, trajectory consistency score, and quality label, and then performing the evidence fusion operator. It is used to determine whether the head-end judgment is sufficiently supported under the current imaging and trajectory evidence. The base interference intensity is represented by normalizing the motion intensity score, occlusion coverage ratio, and link quality penalty term, and then synthesizing them into a single interference intensity using an interference fusion operator. The value range is specified as a base disturbance level caused solely by motion, occlusion, and link anomalies. The evidence contradiction penalty factor is calculated by measuring the inconsistency between the trajectory continuity score and the trajectory consistency score, the mismatch between the head-end credibility assessment value and the observable quality item of the imaging, and the deviation between the head-end credibility and the trajectory evidence. The inconsistency is then mapped to a monotonically decreasing penalty factor with a value ranging from zero to one. It is used to actively reduce the strength of evidence when there is mutual contradiction. The conditional gating inhibition factor is obtained by thresholding or continuously gating the motion intensity score and the occlusion coverage ratio. Its value ranges from zero to one and is used to softly suppress the evidence link under high motion or high occlusion conditions. The interference coupling value is represented by calculating the coupling amount between motion and blockage, the coupling amount between blockage and link anomaly, and the coupling amount between motion and link anomaly. The coupling amount is then mapped to a monotonically increasing amplification intensity, which is used to actively amplify the overall interference intensity on the denominator side when interference occurs concurrently. The time fluctuation penalty value is represented by statistically analyzing the changes in the head-end reliability assessment value, trajectory continuity score, and trajectory consistency score of adjacent uniformly sorted time slices, and mapping the changes to a monotonically increasing fluctuation penalty intensity, which is used to make the score more sensitive to discontinuities and jitter. This represents a minimal positive real constant, obtained by giving a minimal positive real constant and keeping it unchanged within the session. Its value range is greater than zero and much less than one, used to avoid the logarithm becoming uncomputable when the numerator or denominator is zero. This represents a monotonically bounded mapping function, implemented by selecting a monotonically increasing mapping form with bounded output. It maps the logarithmic ratio to a risk quantity between zero and one, ensuring that the comprehensive risk assessment value of navigation is naturally bounded and directly gated, and that the sensitivity of the risk output can be adjusted through shape parameters. This represents the evidence shaping operator, used to perform scaling, truncation, smoothing filtering, and nonlinear shaping on the input evidence product term. This represents the interference aggregation operator, used to perform nonlinear aggregation and amplitude limiting shaping on the basic interference strength, interference coupling value, and time fluctuation penalty value.

[0141] Furthermore, when the comprehensive navigation risk assessment value is not higher than the secondary risk threshold, the system is determined to be in normal navigation mode and outputs a control operation prompt and a scan prompt, while maintaining the default configurations for the gating radius, bridging window length, and smoothing parameters. When the comprehensive navigation risk assessment value is between the secondary and primary risk thresholds, a prompt degradation strategy is triggered, downgrading the control operation prompt to a scan-only prompt and widening the gating range to increase the recapture probability, while extending the sliding time window to suppress transient jumps. When the comprehensive navigation risk assessment value is higher than the primary risk threshold or several consecutive frames are in an unavailable state, a session protection strategy is triggered, freezing the headend output and direction guidance, retaining only the quality alarm and window recapture prompt, and writing the trigger reason code, trigger window number, and keyframe index to the audit log to ensure replayability. The self-adaptive control results are output. The self-adaptive control results include: the current navigation mode flag, gating range parameters, bridging window length, smoothing parameters, prompt level, recapture command, and control records bound to the unified sorting time.

[0142] In this embodiment of the invention, by conducting a comprehensive risk assessment of the navigation process and triggering a closed-loop self-adaptation and degradation strategy, the navigation mode is dynamically adjusted based on the real-time comparison of evidence and interference, thereby achieving adaptive protection and continuous availability of the system.

[0143] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: In this invention, harmonic selective extraction of contrast images effectively suppresses interference from the linear fundamental wave of tissue and broadband strong echoes from gas reflections, highlighting the nonlinear response of microbubbles and significantly enhancing the separability of tubes from artifacts such as blood vessels, intestinal walls, and gas. This overcomes the shortcomings of relying on operator subjective interpretation and easy confusion in tube identification under complex sonographic environments. Simultaneously, by constructing a spatiotemporal map of candidate tubes and performing cross-frame evidence aggregation and topological connectivity constraints, trajectory continuity can be maintained even when tubes are obscured or unevenly visualized, overcoming the trajectory chain breakage caused by single-frame judgment. This achieves stable tube tracking and continuous head-end positioning, meeting the clinical demand for high-precision, high-reliability intelligent navigation.

[0144] Reference manual attached Figure 3 The diagram shows a schematic of the structure of an intelligent navigation system for jejunal catheter placement based on real-time ultrasound contrast imaging provided by the present invention.

[0145] The present invention also provides a jejunal tube placement intelligent navigation system 20 based on real-time ultrasound contrast imaging, applied to the above-mentioned jejunal tube placement intelligent navigation method based on real-time ultrasound contrast imaging, comprising: Processor 201.

[0146] The memory 202 stores computer-readable instructions. When the computer-readable instructions are executed by the processor 201, the intelligent navigation method for jejunal catheter placement based on real-time ultrasound contrast imaging, as described in the method embodiment, is implemented.

[0147] The intelligent navigation system 20 for jejunal tube placement based on real-time ultrasound contrast imaging provided by this invention can execute the above-mentioned intelligent navigation method for jejunal tube placement based on real-time ultrasound contrast imaging and achieve the same or similar technical effects. To avoid repetition, this invention will not elaborate further.

[0148] It should be understood that the processor in the embodiments of the present invention can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor.

[0149] It should also be understood that the memory in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0150] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0151] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0152] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0153] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0154] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0155] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0156] In the embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0157] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0158] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0159] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0160] This invention provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the intelligent navigation method for jejunal catheter placement based on real-time ultrasound contrast imaging as described in the method embodiment.

[0161] The present invention provides a computer-readable storage medium that can implement the steps and effects of the intelligent navigation method for jejunal catheter placement based on real-time ultrasound contrast imaging in the above-described method embodiments. To avoid repetition, the present invention will not elaborate further.

[0162] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0163] The following points need to be explained: (1) The accompanying drawings of the embodiments of the present invention only involve the structures involved in the embodiments of the present invention. Other structures can refer to the general design.

[0164] (2) For clarity, the thickness of layers or regions is enlarged or reduced in the drawings used to describe embodiments of the invention, i.e., these drawings are not drawn to scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being “above” or “below” another element, the element may be “directly” located “above” or “below” the other element or there may be intermediate elements.

[0165] (3) Where there is no conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other to obtain new embodiments.

[0166] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A smart navigation method for jejunal catheter placement based on real-time ultrasound contrast imaging, characterized in that, include: S1: Collect multiple datasets; S2: Perform unified time base alignment on the multiple datasets, and perform structural encapsulation and normalization processing to obtain jejunal tube placement navigation acquisition data frames; S3: Perform harmonic selective extraction on the contrast frames in the jejunal tube placement navigation acquisition data frames to obtain a harmonic selective enhancement result set; S4: Based on the harmonic selective enhancement result set, generate a tube probability map to obtain a tube candidate dataset; S5: Based on the candidate tube dataset, construct a spatiotemporal graph of the candidate tubes and generate a consistent tube trajectory set; S6: Perform head-end positioning decision and reliability assessment on the consistent tube trajectory set to determine the jejunal tube head-end positioning result; S7: Based on the positioning result of the jejunal tube tip, generate navigation prompt information and output navigation interaction results; S8: Perform a comprehensive navigation risk assessment on the navigation interaction results, and trigger a closed-loop self-adaptation and degradation strategy based on the assessment results.

2. The intelligent navigation method for jejunal catheter placement based on real-time ultrasound contrast imaging according to claim 1, characterized in that, The various datasets specifically include: real-time B-mode ultrasound image dataset, real-time ultrasound contrast image dataset, harmonic and nonlinear enhancement feature dataset, imaging state parameter dataset, contrast agent preparation and injection event dataset, catheter placement operation context dataset, probe scanning window position and attitude dataset, and channel identifier dataset.

3. The intelligent navigation method for jejunal catheter placement based on real-time ultrasound contrast imaging according to claim 1, characterized in that, S2 specifically includes: S201: Classify, label, and prioritize the multi-class datasets, and align the multi-class datasets to time using a unified time base through an edge acquisition gateway; S202: Construct data frames from the aligned multi-class datasets to generate the jejunal tube placement navigation acquisition data frames; S203: Through the edge acquisition gateway, maintain the global acquisition queue and the local queue, and perform queue sorting and buffering of the jejunal catheterization navigation acquisition data frames; wherein, the global acquisition queue is sorted according to a unified sorting time, and the local queue is divided according to channel and frame type; S204: Perform linear scaling and normalization processing on the image data in the jejunal tube placement navigation acquisition data frame; S205: Based on the processing results, dynamic range compression and background baseline correction are performed on the intensity of the contrast image to form a standardized enhancement intensity metric consistent with harmonic selective extraction; S206: Based on the standardized enhanced strength metric, the quality marker field is thresholded and hierarchically encoded to mark each data frame as either available or unavailable.

4. The intelligent navigation method for jejunal catheter placement based on real-time ultrasound contrast imaging according to claim 1, characterized in that, S3 specifically includes: S301: Perform mode splitting, windowed recombination, and inter-frame registration on the data frame sequence of the jejunal catheter placement navigation acquisition data frame to obtain the contrast processing input set; wherein, the contrast processing input set includes a subset of contrast frames, a subset of B-mode reference frames, a subset of harmonic and nonlinear enhancement features, a subset of imaging state parameters, a subset of contrast stage state machines, and a subset of quality markers; S302: Using the B-mode reference frame subset as the structural reference, estimate the inter-frame displacement field through phase correlation; S303: Apply the inter-frame displacement field to the angiography frame subset and the harmonic feature subset to complete the pose registration and obtain the motion-disrupted enhanced sequence; S304: Based on the enhancement sequence, an initial harmonic selective enhancement result set is generated by selectively extracting harmonics and suppressing linear fundamental frequencies. S305: Based on the initial harmonic selective enhancement result set, construct the tissue echo baseline using the B-mode structure channel in the B-mode reference frame subset, and perform baseline subtraction on the contrast enhancement channel to obtain the enhancement residual map; S306: Apply edge-preserving filtering and slender structure enhancement operators to the enhanced residual map, and output the harmonic selective enhancement result set.

5. The intelligent navigation method for jejunal catheter placement based on real-time ultrasound contrast imaging according to claim 1, characterized in that, S4 specifically includes: S401: Based on the inter-frame displacement field in the harmonic selective enhancement result set, the harmonic selective enhancement map, enhancement residual map, enhancement direction consistency map, enhancement persistence map, and B-mode reference structure channel are resampled to the same pixel coordinate system to obtain the registered multi-channel feature set; S402: Perform scale normalization and dynamic range compression processing on each channel of the multi-channel feature set, construct a condition vector formed by imaging state parameters, and perform conditional weighted fusion on each channel through the condition vector to form a fused feature tensor; S403: Perform channel mixing on the fused feature tensor and extract local spatial features, output the local spatial features to the segmentation head or detection head to obtain the tube body probability map; S404: Based on the tube probability map, extract the candidate connected component set and the candidate centerline seed set to generate an initial tube candidate dataset; S405: Calculate the quality score for each candidate connected component in the initial tube candidate dataset; wherein, the calculation method for the quality score is as follows: The average tube confidence score is obtained by summing the tube probability values ​​of each pixel in the candidate region of the initial tube candidate dataset and dividing the sum by the total number of pixels in the candidate region. The slenderness index of the candidate region is multiplied by the slenderness reward coefficient and then an exponential operation is performed to obtain the slenderness reward item. Multiply the curvature change rate index of the centerline of the candidate region by the curvature penalty coefficient, take the negative value, and then perform an exponential operation to obtain the curvature penalty term. Multiply the occlusion coverage ratio index by the occlusion penalty coefficient, take the negative value, and then perform an exponential calculation to obtain the occlusion penalty term. The quality score is obtained by multiplying the average tube confidence, the elongation reward, the curvature penalty, and the occlusion penalty together. S406: Based on the quality score and a quality threshold, the initial candidate pipeline dataset is subjected to hierarchical screening; wherein, the hierarchical screening specifically involves: Candidates whose quality score is lower than the quality threshold are marked as first-level candidates and eliminated or downweighted; candidates whose quality score is not lower than the quality threshold are considered as valid candidates, and are sorted from high to low according to their quality score, and the first preset number of candidates are selected as main candidates. S407: Perform morphological and physical prior constraint filtering on the filtered candidate tube connected component set and the candidate centerline seed set, and output the tube candidate dataset.

6. The intelligent navigation method for jejunal catheter placement based on real-time ultrasound contrast imaging according to claim 1, characterized in that, S5 specifically includes: S501: Based on the candidate dataset of the tube body, using the unified sorting time as the index, extract the candidate set of the previous frame and the candidate set of the next frame from the candidate dataset of the tube body, and establish candidate index tables respectively. S502: Based on the candidate index table, construct a cross-frame cost matrix for each pair of candidates in adjacent frames; wherein each element in the cross-frame cost matrix represents the association cost between the candidate in the previous frame and the candidate in the next frame. S503: Based on the cross-frame cost matrix, perform one-to-one matching on the candidates using the minimum cost matching strategy; wherein, the matching rule is as follows: a candidate at the same time is only allowed to be matched with a candidate at the next time, and a candidate at the same time is only allowed to be matched with a candidate at the previous time. S504: Write the successfully matched candidate pairs as directed edges into the edge set, and write the edge attributes such as cost, prediction error, direction consistency, shape similarity and occlusion penalty for each edge to form a candidate tube spatiotemporal graph. S505: Based on the spatiotemporal graph of the candidate tube, spatiotemporal connectivity constraints are applied to the candidate tube to generate the consistent tube trajectory set.

7. The intelligent navigation method for jejunal catheter placement based on real-time ultrasound contrast imaging according to claim 1, characterized in that, S6 specifically includes: S601: Based on the consistent tube trajectory set and the harmonic selective enhancement result set, extract the local evidence vectors of the two endpoints of the main trajectory in each frame or each sliding time window. S602: Based on the local evidence vector, calculate the endpoint cost value of each endpoint of the main trajectory; wherein, the calculation method of the endpoint cost value is as follows: Based on the mean tube probability, mean harmonic enhancement, enhancement persistence score, anchor point consistency score, and candidate quality score in the local evidence vector, the first type of squared deviation term is obtained by squaring the deviation by subtracting one from each of them. The occlusion coverage ratio and position jitter are squared respectively to obtain the second type of squared deviation term; The first type of squared deviation term and the second type of squared deviation term are weighted and summed to obtain the endpoint cost; S603: By using the temperature parameter, the endpoint cost is mapped to the probability of the endpoint head, thus obtaining the head-end reliability assessment value; wherein, the calculation method of the head-end reliability assessment value is as follows: Divide the endpoint value of the first endpoint by the temperature parameter, take the negative value, and then perform an exponential operation to obtain the exponential weight of the first endpoint. Divide the endpoint value of the second endpoint by the temperature parameter, take the negative value, and then perform an exponential operation to obtain the exponential weight of the second endpoint. Add the first endpoint index weight to the second endpoint index weight to obtain the sum of the endpoint index weights; Divide the endpoint value of any endpoint by the temperature parameter, take the negative value, and then perform an exponential operation to obtain the exponential weight of that endpoint. Divide the exponential weight of the endpoint by the sum of the exponential weights to obtain the confidence assessment value of the head end; S604: The endpoint with the largest confidence evaluation value is determined as the head endpoint, and the coordinates of the endpoint are output as the head coordinates, and the tangential direction of the centerline at the endpoint is output as the head direction vector. S605: In conjunction with the confidence threshold, when the head-end confidence assessment value is lower than the confidence threshold or the endpoint jitter is higher than the jitter upper limit threshold, the head-end status is marked as restricted or unavailable, and a navigation degradation strategy is triggered. S606: Output the jejunal tube placement cephalic position result, which includes cephalic position coordinates, cephalic position orientation vector, cephalic position confidence assessment value, and cephalic position status marker.

8. The intelligent navigation method for jejunal catheter placement based on real-time ultrasound contrast imaging according to claim 1, characterized in that, Specifically, S7 includes: S701: Based on the set of jejunal tube tip positioning results, the set of harmonic selective enhancement results, and the tube body probability map, the unified sorting time in the jejunal tube navigation acquisition data frame is used as the unique time index. S702: Based on the unified sorting time, the head-end positioning results, trajectory endpoint sequences, window labels, harmonic selective enhancement maps, tube probability maps, occlusion masks, and quality marker fields under the same time index are aligned and jointly retrieved to obtain multimodal navigation data; S703: Within each time slice corresponding to a unified sorting time, based on the multimodal navigation data, read the head-end coordinates and head-end direction vector of the current time slice, and map the head-end coordinates and head-end direction vector to the pixel coordinate system of the current display frame; S704: Combining the headend status marker, occlusion coverage ratio, and link quality field of the current time slice, when the headend status marker is restricted or unavailable, or the occlusion coverage ratio exceeds the occlusion limit, or the link quality field indicates the presence of a frame loss filling marker and a timestamp abnormality marker, the prompt will be downgraded to only outputting a scan adjustment prompt and a quality alarm prompt, and the placement operation prompt and headend direction guidance will be frozen. S705: Using the unified sorting time as the time granularity, output the navigation interaction results time slice by time slice; wherein, the navigation interaction results include the overlay guidance layer and the prompt level corresponding to the time slice.

9. The intelligent navigation method for jejunal catheter placement based on real-time ultrasound contrast imaging according to claim 1, characterized in that, S8 specifically includes: S801: Input the head-end reliability assessment value, head-end status marker, trajectory continuity score, and trajectory consistency score as input parameters for the comprehensive navigation risk assessment; S802: Calculate the basic evidence strength and basic interference strength based on the input parameters; S803: Multiply the basic evidence strength, evidence contradiction penalty factor and conditional gating inhibition factor to obtain the evidence synthesis quantity; S804: Input the evidence synthesis amount into the evidence shaping operator to obtain the evidence-side output; S805: Input the basic interference strength, interference coupling value and time fluctuation penalty value into the interference aggregation operator to obtain the interference-side output; S806: Add a very small positive real constant to the evidence side output and the interference side output respectively, and divide the processed interference side output by the processed evidence side output to obtain the interference term; S807: After taking the natural logarithm of the interference term, input the monotonically bounded mapping function to obtain the comprehensive navigation risk assessment value; S808: Compare the comprehensive navigation risk assessment value with different risk thresholds. When the comprehensive navigation risk assessment value is not higher than the secondary risk threshold, it is determined that the navigation is in normal mode, and a management operation prompt and a scanning prompt are output. When the comprehensive navigation risk assessment value is between the secondary risk threshold and the primary risk threshold, a downgrade strategy is triggered. When the comprehensive navigation risk assessment value is higher than the primary risk threshold or the system is in an unavailable state for several consecutive frames, a session protection strategy is triggered. S809: Output adaptive control results; wherein, the adaptive control results include navigation mode flags and strategy control parameters corresponding to the navigation mode.

10. A smart navigation system for jejunal catheter placement based on real-time ultrasound contrast imaging, characterized in that, include: processor; A memory storing computer-readable instructions, which, when executed by the processor, implement the intelligent navigation method for jejunal catheter placement based on real-time ultrasound contrast imaging as described in any one of claims 1 to 9.

Images