A Method and System for Estimating the Insertion Length of Infusion Port Catheters Based on Medical Image Recognition

Abstract

This invention discloses a method and system for estimating the insertion length of infusion port catheters based on medical image recognition, specifically relating to the field of medical information processing. It addresses the problem of periodic tip drift during existing infusion port catheter insertion. The method includes: acquiring enhanced CT image sequences covering the superior vena cava to the posterior wall of the atrium; extracting continuous edge attenuation regions; constructing a set of frame-level fuzzy boundary contour curves; performing phase calibration on the image frames using cardiac cycle signals; generating phase-stable contour clusters through phase clustering; and interpolating and fitting the contour point trajectories to generate a fuzzy domain deformation envelope map reflecting the maximum deformation range of the boundary. Further, the drift trajectory of the catheter tip within the envelope map is extracted, a tip drift path cluster is constructed, high-frequency clustering regions are identified through spatial density estimation, stable intervals are determined, and the extension length range of each path within the interval is calculated, outputting the effective insertion length range of the catheter.

Description

Technical Field

[0001] This invention relates to the field of medical information processing technology, and more specifically, to a method and system for estimating the insertion length of infusion port catheters based on medical image recognition. Background Technology

[0002] During the insertion of central venous access catheters, the accuracy of the catheter tip's position directly affects the safety and therapeutic efficacy of drug delivery. Currently, clinical practice generally relies on enhanced CT or X-ray images to determine the static position of the catheter tip, typically using whether the tip reaches the junction of the superior vena cava and the right atrium as the acceptance criterion. However, this method ignores the periodic drift of the catheter tip caused by physiological activities such as respiration and heartbeat, especially significant in the superior vena cava inlet region. Furthermore, due to the dense structure in this area, static images often suffer from blurred boundaries and signal attenuation, further exacerbating catheter positioning errors.

[0003] In some patients, even if intraoperative imaging shows the catheter is in the correct position, postoperative arrhythmias or catheter dysfunction may still occur due to tip displacement into the atrium. This reflects the lack of quantitative support for the "catheter drift tolerance range" in current estimation methods. In particular, there is a lack of systematic modeling and utilization of image signal-to-noise fluctuations in the blurred areas of the superior vena cava. Currently, there is no solution to establish a catheter drift tolerance model based on the characteristics of such blurred imaging areas and to deduce the optimal insertion length in reverse, indicating a significant gap in technical approaches. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a method for estimating the insertion length of infusion port catheters based on medical image recognition to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] The method for estimating the insertion length of infusion port catheters based on medical image recognition includes the following steps:

[0007] S1: Obtain the enhanced CT image sequence of the superior vena cava inlet segment and extract the continuous edge attenuation region in the image sequence;

[0008] S2: Construct a set of temporally fuzzy boundary contour change curves based on continuous edge decay regions;

[0009] S3: Acquire heartbeat signals and calibrate the heartbeat phase to which the image sequence belongs; classify and integrate the set of blurred boundary contour change curves through phase clustering.

[0010] S4: Interpolate and fit the fuzzy boundary contours within the phase cluster to generate a fuzzy domain deformation envelope map under the phase-stable morphology.

[0011] S5: Extract the drift trajectory of the catheter tip within the fuzzy domain deformable envelope map and construct a cluster of catheter tip drift paths;

[0012] S6: Construct a stable interval zone at the tip position based on the drift path cluster, and calculate the effective insertion length range of the catheter within this stable interval zone.

[0013] In a preferred embodiment, in S1, acquiring an enhanced CT image sequence of the superior vena cava inlet segment, and extracting continuous edge attenuation regions in the image sequence specifically includes:

[0014] Acquire enhanced CT image sequences covering the space from the superior vena cava to the posterior wall of the atrium;

[0015] Based on the anatomical location of the superior vena cava entrance, locate the set of cross-sectional frames containing the superior vena cava inlet segment in the image sequence;

[0016] Multi-scale directional convolution is performed on the set of cross-sectional frames of the incoming segment to extract the edge response map between the superior vena cava lumen and adjacent tissues;

[0017] Identify regions where edge intensity decreases continuously in the edge response map and construct a candidate mask set for continuous edge attenuation bands;

[0018] Spatial connectivity filtering is performed on the candidate mask set to exclude non-closed structural regions and retain structurally continuous closed regions as edge attenuation regions.

[0019] In a preferred embodiment, in S2, constructing a set of temporally blurred boundary contour change curves based on continuous edge attenuation regions specifically includes:

[0020] The boundary of the edge attenuation region is projected onto a two-dimensional plane coordinate system to generate a frame-level boundary point set, and the boundary points at the same spatial location are paired up according to the frame index order.

[0021] Track the movement path of the boundary point over time at each corresponding spatial location, and record the boundary drift trajectory with a coordinate change sequence.

[0022] Spline interpolation is performed on all boundary drift trajectories to generate a continuous set of temporally fuzzy boundary profile change curves.

[0023] In a preferred embodiment, in S3, acquiring the heartbeat signal and identifying the heartbeat phase to which the image sequence belongs, and classifying and integrating the set of blurred boundary contour change curves through phase clustering, specifically includes:

[0024] The cardiac cycle signal acquired based on continuous electrical signal is subjected to discrete-time mapping processing to generate a frame-level cardiac phase tag sequence.

[0025] Among them, the heartbeat phase refers to a specific time segment obtained by dividing a complete heartbeat cycle into time segments according to a preset division rule;

[0026] Construct a mapping structure between the set of fuzzy boundary contour change curves and the frame-level heartbeat phase label sequence, rearrange the frame order of the fuzzy boundary contour change curves, and classify all contour change curve frames into the phase label set.

[0027] Based on the similarity of the spatial distribution of contour boundary points, the frame sequence of fuzzy boundary contour change curves under the same phase label is used to determine the similarity of the contour boundary point frame sequence. Unsupervised clustering is used to generate a cluster of phase-stable contour curves, which serve as the fuzzy boundary contours within the phase cluster.

[0028] In a preferred embodiment, the step of performing frame-sequence discrimination based on the spatial distribution morphology similarity of contour boundary points on the fuzzy boundary contour change curves under the same phase label, and generating a cluster of phase-stable contour curves through unsupervised clustering, specifically includes:

[0029] Uniform sampling of contour boundary points is performed on each fuzzy boundary contour curve under the same phase label to construct a set of point sequences with a unified dimension;

[0030] For each contour boundary point sequence, calculate the Euclidean distance change sequence and angle change sequence between adjacent points, and extract the geometric feature trajectory of the spatial boundary change;

[0031] Based on geometric feature trajectories, dynamic time warping is performed on all contours to obtain the morphological deformation distance matrix between any two contour curves.

[0032] Hierarchical clustering analysis is performed on the distance matrix, and the contour curves are grouped according to the preset variation difference threshold to generate similar morphological clusters as phase-stable contour curve clusters.

[0033] In a preferred embodiment, in S4, interpolating and fitting the fuzzy boundary contours within the phase cluster to generate a fuzzy domain deformation envelope map under a phase-stable morphology specifically includes:

[0034] Extract the set of boundary points from each phase-stable contour curve cluster, and relocate the starting point of each contour curve according to a unified coordinate direction;

[0035] Arrange the boundary points in the contour curve cluster according to the order of their corresponding positions in the frame sequence to construct a multi-frame boundary point trajectory set;

[0036] For each trajectory in the boundary point trajectory set, spatial curve interpolation fitting is performed to generate a multi-frame continuous deformation path, and the boundary maximum envelope is reconstructed based on the continuous deformation path.

[0037] The boundary curve structure of the maximum envelope frame is closed and processed into a closed three-dimensional surface graphic, which is then defined as the fuzzy domain deformed envelope map under the corresponding heartbeat phase.

[0038] In a preferred embodiment, in S5, extracting the drift trajectory of the catheter tip within the fuzzy domain deformed envelope map and constructing the catheter tip drift path cluster specifically includes:

[0039] In enhanced CT image sequences, the edge bundles of the catheter structure are identified, and image frame extraction is performed on the coordinates of the end point in each frame to generate a sequence of catheter tip points.

[0040] A three-dimensional spatial remapping is applied to each endpoint in the duct tip sequence to filter out the set of tips located inside the phase deformation envelope execution profile;

[0041] Based on the screened tip point set, reconstruct the spatial path lines arranged in ascending order of frame index, and connect them on the time axis to form the duct tip drift trajectory line;

[0042] An angle transformation rate sequencing process is applied to the drift trajectory line to segment the trajectory segments with continuous directional trends in the path curvature sequence and generate a set of trajectory structure fragments.

[0043] Based on the similarity of point sets in the spatial distribution of each trajectory structure segment, a merging operation is performed to classify trajectory segments that meet the criteria of continuity and distribution consistency into a unified path cluster, and output the duct tip drift path cluster.

[0044] In a preferred embodiment, in S6, constructing a stable interval zone at the tip position based on the drift path cluster, and calculating the effective insertion length range of the catheter within this stable interval zone specifically includes:

[0045] Spatial density estimation is performed on the tip endpoints of all paths in the drift path cluster to construct a probability distribution map of the tip clustering region;

[0046] In the probability distribution map, spatially enclosed regions with continuous distribution density higher than a set threshold are extracted and defined as candidate regions of the apex stable interval zone.

[0047] The projected distance of the intersection point between each path in the drift path cluster and the boundary of the stable interval zone is calculated and used as the line segment extension length of the corresponding path within the stable interval zone.

[0048] Merge the upper and lower boundaries of all line segment extension lengths to generate a set of insertion depth length intervals for the catheter tip within the stable interval zone.

[0049] On the other hand, the present invention provides a system for estimating the insertion length of an infusion port catheter based on medical image recognition, including an image extraction module, a contour change construction module, a contour change clustering module, a deformation envelope map generation module, and a stable insertion estimation module:

[0050] Image extraction module: Acquire enhanced CT image sequences covering the superior vena cava to the posterior wall of the atrium, locate cross-sectional frames containing the superior vena cava confluence segment, perform multi-scale convolution on the frame set to extract edge response maps, identify gray-scale continuous attenuation regions, and construct edge attenuation regions by filtering through spatial connectivity;

[0051] Contour Change Construction Module: Projects the boundary of the edge attenuation region onto a two-dimensional coordinate system to generate a frame-level boundary point set, tracks the change path of the points over time and interpolates, and constructs a set of fuzzy contour curves that reflect the dynamic changes of the boundary.

[0052] Contour variation clustering module: It establishes a mapping between contour variation curves and heartbeat phase labels, classifies and clusters contour sequences according to labels, and performs unsupervised clustering based on spatial morphological similarity to generate a cluster of phase-stable contour curves.

[0053] Deformation envelope map generation module: Extracts the boundary point trajectory of stable contour clusters, performs interpolation fitting to reconstruct the deformation path, constructs the boundary maximum envelope structure, and generates a three-dimensional closed figure as the fuzzy domain deformation envelope map;

[0054] Stable insertion estimation module: Extracts catheter tip sequence and filters trajectories falling within the envelope map, identifies and clusters tip drift path segments, constructs stable interval bands by combining spatial density distribution, and outputs the effective insertion length range.

[0055] The technical effects and advantages of the present invention regarding the method and system for estimating the insertion length of infusion port catheters based on medical image recognition are as follows:

[0056] By extracting the continuous edge attenuation region of the superior vena cava inlet segment, and combining the construction of temporal fuzzy boundary contour change curves with dynamic calibration of cardiac phase, the spatial drift behavior of the catheter tip under physiological fluctuations is modeled. By interpolating and fitting the phase clustering contours, a fuzzy domain deformable three-dimensional envelope map is generated, effectively characterizing the possible activity boundaries of the catheter. Furthermore, by combining the structural clustering of the catheter tip drift trajectory, path clusters are constructed and stable intervals are identified, thereby deriving an individualized effective catheter insertion length range. This approach has the technical advantages of strong dynamic adaptability, low anatomical dependence, and high intraoperative guidance value. Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating the method for estimating the insertion length of an infusion port catheter based on medical image recognition according to the present invention.

[0058] Figure 2This is a schematic diagram of the infusion port catheter insertion length estimation system based on medical image recognition according to the present invention. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0060] Example 1

[0061] Figure 1 The present invention provides a method for estimating the insertion length of an infusion port catheter based on medical image recognition, which includes the following steps:

[0062] S1: Obtain the enhanced CT image sequence of the superior vena cava inlet segment and extract the continuous edge attenuation region in the image sequence;

[0063] S2: Construct a set of temporally fuzzy boundary contour change curves based on continuous edge decay regions;

[0064] S3: Acquire heartbeat signals and calibrate the heartbeat phase to which the image sequence belongs; classify and integrate the set of blurred boundary contour change curves through phase clustering.

[0065] S4: Interpolate and fit the fuzzy boundary contours within the phase cluster to generate a fuzzy domain deformation envelope map under the phase-stable morphology.

[0066] S5: Extract the drift trajectory of the catheter tip within the fuzzy domain deformable envelope map and construct a cluster of catheter tip drift paths;

[0067] S6: Construct a stable interval zone at the tip position based on the drift path cluster, and calculate the effective insertion length range of the catheter within this stable interval zone.

[0068] In S1, an enhanced CT image sequence of the superior vena cava inlet segment is acquired, and continuous edge attenuation regions in the image sequence are extracted.

[0069] Enhanced CT data acquisition was performed, with the acquisition range set from the mid-segment of the superior vena cava to the posterior wall of the right atrium to fully cover the expected activity segment of the catheter tip. A 64-slice spiral CT scanner was used for image acquisition, and a non-ionic contrast agent (iohexol) was injected. The dosage and flow rate were automatically adjusted according to the patient's weight. Acquisition was performed during a resting chest scan. The image reconstruction resolution was set to 0.5mm × 0.5mm on the horizontal axis, with a slice thickness of 1mm, and the total number of image frames was controlled between 80 and 150 frames.

[0070] After acquisition, a frame-level image sequence matrix is ​​constructed. All images undergo preprocessing, including window width and level normalization, noise filtering (using median filtering), and postural correction (based on skeleton registration). Structural localization is then performed. Combining baseline anatomical features such as the ascending aorta, right atrial wall, and intercostal space distribution, the index interval of the junction between the terminal superior vena cava and the right atrium in the image sequence is calculated. Using center-point trajectory backtracking technology, the vascular path is traced downwards from the mid-segment SVC, ultimately locating a set of cross-sectional frames containing the superior vena cava's entry into the atrium, typically 20–30 frames, referred to as the analysis frame set.

[0071] For this frame set, multi-scale directional edge convolution operations are performed on each frame. First, derivative convolution kernels with different directions are constructed based on a Gaussian function, including directions of 0°, 45°, 90°, and 135°. The kernel size is adjusted to three levels: 3×3, 5×5, and 7×7 according to a scale-space strategy. The direction of maximum response for each pixel is selected from the convolution result to form a direction-enhanced edge response map. Subsequently, gray-level scan lines are generated in the edge response map with the radial direction of the blood vessel as the direction, the gray-level change curve is extracted, and the first derivative is calculated. When the gray-level intensity of multiple consecutive pixels in a certain segment decreases, and the derivative value is stably within a set slope threshold range, it is considered an edge intensity attenuation region.

[0072] All pixels that meet the attenuation conditions are defined as candidate attenuation masks, forming the initial set of edge attenuation bands for that frame. A spatial connectivity analysis method is introduced to perform 8-neighborhood connectivity checks on the candidate mask set, identifying closed structures and eliminating regions with broken boundaries, insufficient area, or no connectivity. Finally, closed edge attenuation regions with intact structures surrounding the duct space are retained. The closed region output for each frame is a binary mask image, and its center trajectory and boundary coordinate set are stored.

[0073] In S2, a set of temporally fuzzy boundary contour change curves is constructed based on the continuous edge decay region.

[0074] The closed boundaries of the edge attenuation region in each frame are output as a set of pixel coordinates. To unify the reference system positions of the boundary points in space, all boundary point sets are uniformly projected to a two-dimensional plane coordinate system (XY plane under the Axial view) registered with the cross-section of the CT image. The boundary point set of each frame is labeled as a frame-level boundary point set and stored in an ordered structure with the frame index as the label.

[0075] Based on the temporal correspondence of points, boundary points at the "same spatial location" are found across all frames. A uniform polar angle sampling method is employed, using the centroid coordinates of the edge region of each frame as the polar coordinate center. The boundary curve is resampled into N uniformly distributed polar angle sampling points (e.g., N=360, one point per degree). The boundary contour of each frame is then represented as a sequence of point sets with uniform point numbers; consistent numbers indicate consistent polar angles, i.e., consistent approximate spatial locations.

[0076] In all frames, points with the same index are paired to construct their trajectories as they change with frame sequence (or time). For example, a boundary point with a frame number corresponding to a specific location has a specific coordinate position in each frame of the image sequence. By arranging these coordinates sequentially according to the time order of the frame index, a sequence of the boundary point's movement trajectory over time is formed, representing its positional evolution throughout the entire image sequence. This sequence reflects the positional changes of the boundary location due to heartbeats or other physiological factors throughout the image sequence, i.e., the boundary drift trajectory.

[0077] For each boundary point drift trajectory, further fitting and smoothing processing is performed. In this embodiment, cubic spline interpolation is used to reconstruct the coordinate changes in the x-axis and y-axis directions respectively. During interpolation, the time index of the boundary points is normalized to ensure that the interpolation process is carried out on a uniform time scale. After interpolation, each trajectory becomes a continuous spatial curve, reflecting the smooth drift behavior of the point throughout the entire frame sequence.

[0078] After completing the trajectory interpolation of all numbered points, the interpolation results of each frame are recombined into a new contour curve. Specifically, for each time point on the interpolated trajectory curve, the corresponding set of points is extracted, and these points are connected sequentially according to their numbering to form a complete frame-level contour line. In this way, a continuous sequence of contour curves is formed on the time axis, with each curve consisting of interpolated points, resulting in higher temporal consistency and smoothness compared to the jump behavior of contour points in the original image.

[0079] In S3, heartbeat signals are acquired and the heartbeat phase to which the image sequence belongs is calibrated. The set of fuzzy boundary contour change curves is classified and integrated through phase clustering.

[0080] While the patient undergoes an enhanced CT scan, their electrocardiogram (ECG) signals are simultaneously acquired. The acquisition device is an ECG monitor with synchronous recording capability, and the sampling frequency is no less than 500Hz. During the acquisition process, the peak point of electrical activity signal at the onset of each heartbeat is automatically detected, and the time interval between each adjacent peak point is defined as a complete heartbeat cycle. The entire image acquisition time usually spans multiple heartbeat cycles, and the system records the division points of these cycles on the time axis. The image frames are matched with the heartbeat rhythm, and the CT device provides precise acquisition time information for each frame. The system matches the acquisition time point of each frame image to its corresponding heartbeat cycle and further calculates the position of the image within the corresponding heartbeat cycle. In this embodiment, each heartbeat cycle is divided into several equally wide time segments, such as based on the QRS onset, T wave peak, and early diastole as the time segment division criteria. Then, based on the relative position of each frame acquisition time within the heartbeat cycle, a fixed phase label is assigned to that frame. After completing the heartbeat phase calibration of all image frames, the frame sequence information with the calibrated phase labels is matched with the previously constructed set of fuzzy boundary contour change curves. Each frame of the image corresponds to a boundary contour curve, which is then assigned to the curve set under the corresponding phase label based on the time index. After this construction is completed, each heartbeat phase segment contains a set of boundary contour curves extracted against a background of similar rhythms.

[0081] For the frame sequence of fuzzy boundary contour change curves under the same phase label, a similarity judgment based on the spatial distribution of contour boundary points is performed. Unsupervised clustering is then used to generate a cluster of phase-stable contour curves, which serve as the fuzzy boundary contours within the phase cluster.

[0082] Point set standardization is performed on each blurred boundary contour curve. Since the number of points on different contour curves in the original image may be inconsistent, and the spacing between discrete points is uneven, this can interfere with subsequent morphological analysis. Therefore, contour point sequence resampling is necessary. This embodiment adopts a polar angle uniform sampling strategy: using the geometric centroid of the contour as the reference center, the contour is projected onto a polar coordinate system, and boundary points are extracted at equal angular intervals, so that each contour is expressed as an ordered set of points with a uniform number of points (e.g., 128 points), and the point sequence number corresponds to a uniform spatial orientation.

[0083] After obtaining a sequence of boundary points with a uniform dimension, the geometric variation features of each contour are extracted. Specifically, the Euclidean distance sequence between each point on the contour and its adjacent points is calculated sequentially to characterize the line segment scale features of the local boundary; simultaneously, the angular variation sequence of the boundary orientation is calculated as a quantitative expression of the curvature. These two types of sequences together constitute a geometric feature trajectory describing the spatial structure changes of the contour, used to measure the deformation relationship between curves.

[0084] To compare the morphological differences between any two contour curves, a dynamic time warping (DTW) method is used to perform pairwise analysis on their geometric feature trajectories. The DTW algorithm allows for elastic alignment of two sets of feature sequences even in the presence of nonlinear point correspondences, thereby obtaining relatively stable morphological deformation distances. After pairwise analysis of all curve pairs, a complete morphological distance matrix between contours is constructed, where matrix elements represent the degree of morphological variation between the two curves.

[0085] After obtaining the distance matrix, hierarchical clustering is used for contour clustering analysis. During the clustering process, curve subsets are merged progressively based on the values ​​of the distance matrix, and the termination condition is determined according to the variational deviation threshold set by the system (specifically set based on the contour construction accuracy, with the default being the average distance). When the maximum morphological distance between contours does not exceed the tolerance range, they are considered to have spatial morphological consistency. Ultimately, multiple contour clusters are formed, each containing several contour curves with similar structures.

[0086] In S4, interpolation fitting is performed on the fuzzy boundary contours within the phase cluster to generate a fuzzy domain deformation envelope map under the phase-stable morphology.

[0087] Boundary point data is extracted from multiple obtained contour curve clusters, with each cluster corresponding to a set of contour curves considered relatively stable in spatial structure. To ensure a boundary structure with a unified spatial orientation, the starting points of all contour curves need to be repositioned. This embodiment employs a contour reparameterization strategy: using the contour centroid as a reference, the direction angle of the line connecting the first point of the contour to the centroid is calculated, and all curves are rotated and translated so that their starting points are in the same direction in the coordinate system. This process ensures that the boundary point positions of different contours have a unified starting point alignment characteristic in time sequence.

[0088] Subsequently, a "boundary point trajectory set" is constructed. Specifically, each contour curve is arranged sequentially according to frame order, and its boundary points are paired according to their numerical order to form a spatial point correspondence across multiple frames. Taking a group of points with the same number as a unit, their spatial positions in different frames are organized along the time axis to generate a point trajectory. Each trajectory represents the deformation path experienced by a fixed boundary position in the time series (i.e., during frame order changes).

[0089] Three-dimensional interpolation is performed on each boundary point trajectory using spline curve interpolation to construct a continuous deformation path for each trajectory point sequence in the spatial dimension. The interpolated trajectory transitions smoothly on the frame axis, reflecting the boundary evolution trend of structural deformation induced by heartbeat. After processing, the interpolation results of all trajectories are aggregated into a three-dimensional coordinate grid, forming a spatial trajectory group describing the outer edge change of the contour under temporal continuity. Based on the outermost edge distribution of each point in the trajectory group in space, the maximum outer shell structure of the point group in three-dimensional space is extracted using convex hull fitting and directional scanning methods. The generated boundary shell is encapsulated into a closed three-dimensional surface graphic using a section closure algorithm, and this structure is defined as the "fuzzy domain deformation envelope map" under the condition of heartbeat phase. The graphic structure encompasses the point trajectories of all stable contour curves and reflects their maximum deformation range. This closed structure is stored using point set grid encoding.

[0090] In S5, the drift trajectory of the catheter tip within the fuzzy domain deformable envelope map is extracted, and a cluster of catheter tip drift paths is constructed.

[0091] An image enhancement method based on grayscale difference and structural refinement is employed, combined with a multi-scale Frangi filter to perform beam enhancement on the image sequence, highlighting the slender, bright areas of the catheter against the vascular background. Region growing and morphological refinement algorithms are used on the processed images to extract the edge beams of the catheter structure, and the farthest endpoint aligned with the image boundary direction in each frame is labeled as the catheter tip position for that frame. The coordinates of the identified catheter tips in each frame are extracted according to the image frame index order to construct a catheter tip point sequence. To ensure that the extracted catheter trajectory truly falls within the physiologically permissible area, each tip point needs to be projected onto the fuzzy domain deformed envelope map generated in the previous step, and a three-dimensional spatial remapping operation is performed. Specifically, the image coordinates (based on pixels) of the tip point are converted into actual spatial coordinates (based on CT image resolution and scanning parameters), and it is checked whether the point is inside the closed surface of the three-dimensional envelope map under the current phase conditions. This embodiment uses a three-dimensional point-mesh containment test algorithm, such as the ray casting method and the centroid triangle method, to jointly determine the point envelope relationship. Only catheter tip points falling inside the envelope map are retained as the basic point set for constructing effective drift trajectories.

[0092] After obtaining a continuous and valid set of catheter tip points, these points are connected in three-dimensional space in ascending order of image frame index to form a directed path, creating the catheter tip drift trajectory. This trajectory represents the actual spatial movement path of the catheter tip in consecutive frames and has temporal continuity. Considering that cardiac activity, respiration, and slight changes in patient position may cause the trajectory to undergo non-unidirectional and stable changes, the morphological changes of this trajectory are further analyzed.

[0093] Specifically, a trajectory curvature analysis mechanism is introduced to calculate the rate of change of angle at each point on the trajectory in space, thereby forming a curvature change sequence. When the trajectory exhibits a trend of directional stability and small angular fluctuation amplitude within a certain continuous range of points, this segment is considered to be a drift segment in the same direction. The entire trajectory curve is then segmented using this method to form a set of multiple trajectory structure segments, each segment representing the stable unidirectional movement behavior of the duct tip over a continuous period of time.

[0094] The trajectory structure segments are merged using the following method: spatial statistical features such as the distribution center, direction vector, and distribution density of each trajectory segment are extracted, and the similarity index between segments is calculated based on the Euclidean distribution overlap and directional consistency between the point sets. Hierarchical clustering is then used to perform spatial clustering analysis on all segments. Based on preset spatial distribution tolerance and directional variation thresholds, trajectory segments that meet the conditions of continuity and distribution consistency are grouped into a unified path cluster. The final output duct tip drift path cluster contains multiple trajectory subclusters. The tip trajectories within each subcluster are continuous in time, exhibit similar trends in space, and possess segmented features structurally suitable for drift modeling and stable interval identification.

[0095] In S6, a stable interval zone at the tip position is constructed based on the drift path cluster, and the effective insertion length range of the catheter within this stable interval zone is calculated.

[0096] In the constructed cluster of catheter tip drift paths, the endpoint coordinates of each path are extracted. Each path endpoint represents the final stable position of the catheter tip over a certain period; therefore, all endpoints in the entire path set constitute a sample of the possible spatial distribution of catheter tips. After aggregating all endpoint coordinates, a 3D point cloud is constructed, and spatial kernel density estimation is performed based on this point set. During kernel density estimation, a 3D Gaussian kernel function is used to smooth the catheter tip endpoint point cloud, forming a spatially continuous density field. The density value represents the probability intensity of tip clustering per unit volume. The kernel bandwidth parameter is adaptively adjusted according to the spatial distribution range of the point cloud to ensure that the estimated probability distribution map retains the structural features of the concentrated region while avoiding the loss of detail caused by over-smoothing. After processing, the probability value of the tip distribution at each location in 3D space is generated, i.e., the probability density map of the tip clustering region.

[0097] In this density map, the system extracts regions with density values ​​higher than a set threshold from continuous areas as candidate regions. The threshold can be set empirically to the quantile of density values ​​above 80% in the overall distribution, or automatically adjusted based on clustering. All spatial locations that meet this density condition constitute a spatially closed region. Through connected component analysis and 3D boundary generation techniques, the geometric boundary of this region is extracted and defined as the "candidate region for stable interval zone of the catheter tip." This region represents the spatial range in which the catheter tip is most likely to remain under dynamic conditions. After completing the extraction of the stable interval zone, it is necessary to evaluate the crossing of each catheter path within this interval. To this end, the intersection point between each path in the drift path cluster and the boundary of the stable interval zone is identified, and the length of the spatial path segment from the path's starting point into the interval zone to its departure from the interval boundary is calculated, defined as the "extension length" of the path within the interval zone. If a path has multiple discontinuous segments within the interval, the longest segment can be retained or the length of the entire segment can be accumulated to enhance the adaptability of the estimation.

[0098] By summarizing and analyzing the extension lengths of all paths within the stable interval zone, the system extracts the maximum and minimum extension lengths as the upper and lower boundaries of the current sample catheter insertion length, thus constructing a "set of effective catheter insertion length intervals." This set can be expressed as a set of length interval values, or further statistically modeled to form a confidence interval expression, serving as a reference for intraoperative catheter advancement depth recommendations. This approach avoids the over-reliance on static anatomical points in traditional insertion depth assessment, instead constructing a dynamic and reliable spatial tolerance zone based on the actual drift behavior of multi-path tips. Combined with trajectory analysis results, it generates individualized depth recommendations, demonstrating significant clinical adaptability, dynamic variability, and modeling interpretability, providing direct data support for intraoperative navigation during port-a-cath implantation.

[0099] Example 2

[0100] The difference between Embodiment 2 and Embodiment 1 is that this embodiment introduces a system for estimating the insertion length of an infusion port catheter based on medical image recognition.

[0101] Figure 2 A schematic diagram of the infusion port catheter insertion length estimation system based on medical image recognition of the present invention is provided, including an image extraction module, a contour change construction module, a contour change clustering module, a deformation envelope map generation module, and a stable insertion estimation module:

[0102] Image extraction module: Acquire enhanced CT image sequences covering the superior vena cava to the posterior wall of the atrium, locate cross-sectional frames containing the superior vena cava confluence segment, perform multi-scale convolution on the frame set to extract edge response maps, identify gray-scale continuous attenuation regions, and construct edge attenuation regions by filtering through spatial connectivity;

[0103] Contour Change Construction Module: Projects the boundary of the edge attenuation region onto a two-dimensional coordinate system to generate a frame-level boundary point set, tracks the change path of the points over time and interpolates, and constructs a set of fuzzy contour curves that reflect the dynamic changes of the boundary.

[0104] Contour variation clustering module: It establishes a mapping between contour variation curves and heartbeat phase labels, classifies and clusters contour sequences according to labels, and performs unsupervised clustering based on spatial morphological similarity to generate a cluster of phase-stable contour curves.

[0105] Deformation envelope map generation module: Extracts the boundary point trajectory of stable contour clusters, performs interpolation fitting to reconstruct the deformation path, constructs the boundary maximum envelope structure, and generates a three-dimensional closed figure as the fuzzy domain deformation envelope map;

[0106] Stable insertion estimation module: Extracts catheter tip sequence and filters trajectories falling within the envelope map, identifies and clusters tip drift path segments, constructs stable interval bands by combining spatial density distribution, and outputs the effective insertion length range.

[0107] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0108] The above embodiments can be implemented, in whole or in part, by software, hardware, 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. The 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 processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the 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. The 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. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0109] Those skilled in the art will recognize that the modules 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 implementation should not be considered beyond the scope of this application.

[0110] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0111] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, 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 apparatuses or modules may be electrical, mechanical, or other forms.

[0112] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0113] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0114] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion 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 described in the various embodiments of this application. 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.

[0115] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0116] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for estimating the length of a catheter placed in a port based on medical image recognition, characterized in that, The method comprises the following steps: S1: acquiring an enhanced CT image sequence of the superior vena cava inlet segment, and extracting a continuous edge attenuation region in the image sequence; S2: constructing a time-series fuzzy boundary profile change curve set based on the continuous edge attenuation region; S3: collecting a heart beat signal and calibrating a heart beat phase to which the image sequence belongs, and classifying and integrating the fuzzy boundary profile change curve set through phase clustering; S4: interpolating and fitting the fuzzy boundary profile in the phase cluster to generate a fuzzy domain deformation envelope graph in a phase stable mode; S5: extracting a drift trajectory of the catheter tip in the fuzzy domain deformation envelope graph to construct a catheter tip drift path cluster; S6: constructing a stable interval band of the tip position according to the drift path cluster, and calculating a corresponding effective insertion length range of the catheter in the stable interval band; In S1, acquiring an enhanced CT image sequence of the superior vena cava inlet segment, and extracting a continuous edge attenuation region in the image sequence specifically comprises: Collecting an enhanced CT image sequence covering the space range from the superior vena cava to the posterior wall of the atrium; According to the anatomical position of the superior vena cava inlet, the cross-sectional frame set containing the superior vena cava inlet segment is positioned in the image sequence; Performing a multi-scale directional convolution operation on the cross-sectional frame set of the inlet segment to extract the edge response graph between the superior vena cava lumen and the adjacent tissue; In S2, constructing a time-series fuzzy boundary profile change curve set based on the continuous edge attenuation region specifically comprises: Projecting the edge attenuation region boundary into a two-dimensional plane coordinate system to generate a frame-level boundary point set, and corresponding pairing the boundary points at the same spatial position in the order of frame index; Tracking the moving path of the boundary point evolution over time at each corresponding spatial position, and recording the boundary drift trajectory in the form of coordinate change sequence; Performing a spline interpolation operation on all boundary drift trajectories to generate a continuous time-series fuzzy boundary profile change curve set; In S3, collecting a heart beat signal and calibrating a heart beat phase to which the image sequence belongs, and classifying and integrating the fuzzy boundary profile change curve set through phase clustering specifically comprises: Discrete time mapping processing is performed on the heart beat cycle signal collected based on the continuous electrical signal to generate a frame-level heart beat phase label sequence; Constructing a mapping structure between the fuzzy boundary profile change curve set and the frame-level heart beat phase label sequence, rearranging the fuzzy boundary profile change curve frame sequence, and grouping all profile change curve frames into phase label sets; In S6, constructing a stable interval band of the tip position according to the drift path cluster, and calculating a corresponding effective insertion length range of the catheter in the stable interval band specifically comprises: Performing spatial density estimation on the tip endpoints of all paths in the drift path cluster to construct a probability distribution graph of the tip aggregation area; Extracting a spatially closed region with a continuous distribution density higher than a set threshold in the probability distribution graph, and defining it as a candidate region of the tip stable interval band.

2. The medical image recognition-based implantable port catheter length estimation method of claim 1, wherein, In S1, acquiring an enhanced CT image sequence of the superior vena cava inlet segment, and extracting a continuous edge attenuation region also includes: Identifying the region with continuous descending edge intensity in the edge response graph to construct a candidate mask set of the continuous edge attenuation band; Performing spatial connectivity screening on the candidate mask set, excluding non-closed structure regions, and retaining closed regions with structure continuity as edge attenuation regions. 3.The medical image recognition-based implantable port catheter length estimation method of claim 1, wherein, In S3, a heartbeat signal is collected and a heartbeat phase to which the image sequence belongs is calibrated, and a set of fuzzy boundary profile change curves is classified and integrated through phase clustering Comprise: Wherein, the heartbeat phase refers to a specific time segment obtained by time axis segmentation of a complete heartbeat cycle according to a preset division rule; The fuzzy boundary profile change curve frame sequence under the same phase label is subjected to similarity discrimination based on the spatial distribution form of the profile boundary points, and a phase stable profile curve cluster is generated through unsupervised clustering as the fuzzy boundary profile in the phase clustering.

4. The medical image recognition-based implantable port catheter length estimation method of claim 3, wherein, The fuzzy boundary profile change curve frame sequence under the same phase label is subjected to similarity discrimination based on the spatial distribution form of the profile boundary points, and a phase stable profile curve cluster is generated through unsupervised clustering, specifically comprising: Uniformly sampling the profile boundary points of each fuzzy boundary profile curve under the same phase label to construct a point sequence set of uniform dimension; Calculate the Euclidean distance change sequence and angle change sequence between adjacent points for each profile boundary point sequence, and extract the geometric feature trajectory of spatial boundary change; Based on the geometric feature trajectory, perform dynamic time warping operation on all profiles to obtain the form deformation distance matrix between any two profile curves; Perform hierarchical clustering analysis on the distance matrix, group the profile curves according to the preset deformation tolerance threshold, and generate a similarity form clustering cluster as the phase stable profile curve cluster. 5.The medical image recognition-based implantable port catheter length estimation method of claim 1, wherein, In S4, the fuzzy boundary profile in the phase clustering is subjected to interpolation fitting to generate a fuzzy domain deformation envelope map under a phase stable form, specifically comprising: Extract the boundary point set in each phase stable profile curve cluster, and reposition the starting points of each profile curve in a uniform coordinate direction; Arrange the boundary points in the profile curve cluster in the order of their corresponding positions in the frame sequence to construct a multi-frame boundary point trajectory set; Perform spatial curve interpolation fitting on each trajectory in the boundary point trajectory set to generate a continuous deformation path between multiple frames, and perform boundary maximum envelope reconstruction based on the continuous deformation path; Close the boundary curve structure of the maximum envelope framework to a closed three-dimensional surface graph, and define the graph as a fuzzy domain deformation envelope map under the corresponding heartbeat phase. 6.The medical image recognition-based implantable port catheter length estimation method of claim 1, wherein, In S5, a catheter tip drift trajectory is extracted in the fuzzy domain deformation envelope map to construct a catheter tip drift path cluster, specifically comprising: Identify the catheter structure edge line bundle in the enhanced CT image sequence, perform image frame extraction on the end point coordinates in each frame to generate a catheter tip point sequence; Apply three-dimensional space remapping to each endpoint in the catheter tip point sequence to screen out a set of tip points located inside the phase deformation envelope map execution profile; Reconstruct a spatial path line arranged in ascending order of frame index based on the screened tip point set, and connect it in the time axis to form a catheter tip drift trajectory line; Apply angle transformation rate sequencing processing to the drift trajectory line to subdivide trajectory sections with continuous direction trend in the path curvature sequence, and generate a trajectory structure fragment set; The trajectory segments meeting the continuity and distribution consistency criteria are classified into a unified path cluster according to the point set similarity of the spatial distribution of each trajectory segment, and a set of catheter tip drift path clusters is output.

7. The medical image recognition-based implantable port catheter length estimation method of claim 1, wherein, In S6, a stable interval band of the tip position is constructed according to the drift path cluster, and the corresponding effective insertion length range of the catheter in the stable interval band is calculated, which further includes: The projection distance of the intersection point of each path in the drift path cluster and the boundary of the stable interval band is calculated as the linear segment extension length of the corresponding path in the stable interval band; The upper and lower boundary merging of all linear segment extension lengths is performed to generate a set of insertion depth length intervals of the catheter tip in the stable interval band.

8. A medical image recognition based on estimation system of length of catheter placement for port, for realizing the medical image recognition based on estimation method of length of catheter placement of any one of claims 1-7, characterized in that, The image extraction module, the contour change construction module, the contour change clustering module, the deformation envelope map generation module, and the stable insertion estimation module are included: The image extraction module: collects enhanced CT image sequences covering the superior vena cava to the atrial posterior wall, locates the cross-sectional frame containing the superior vena cava inflow segment, performs multi-scale convolution to extract edge response maps on the frame set, identifies gray continuous attenuation regions, and constructs edge attenuation regions through spatial connectivity screening; The contour change construction module: projects the edge attenuation region boundary to a two-dimensional coordinate system to generate frame-level boundary point sets, tracks the change path of the point position over time and interpolates, and constructs a set of fuzzy contour curves reflecting the dynamic change of the boundary; The contour change clustering module: maps the contour change curve to the heart beat phase label, classifies and aggregates the contour sequence according to the label, and performs unsupervised clustering based on spatial morphological similarity to generate a stable contour curve cluster; The deformation envelope map generation module: extracts the boundary point trajectory of the stable contour cluster, performs interpolation fitting to reconstruct the deformation path, constructs the maximum envelope structure of the boundary, and generates a three-dimensional closed graph as a fuzzy domain deformation envelope map; The stable insertion estimation module: extracts the catheter tip point sequence and filters the trajectories falling within the envelope map, identifies the tip drift path segments and clusters, constructs a stable interval band in combination with the spatial density distribution, and outputs the effective insertion length range.

Images