Preoperative intelligent planning method for nasal septum mucosa flap pedicle path

Abstract

The application discloses a preoperative intelligent planning method for a nasal septum mucosal flap blood vessel pedicle path, and relates to the technical field of surgical path planning. The method first analyzes the feature difference of physiological structures between the current patient and each historical patient, obtains the reference weight of each historical patient, and further trains a model based on the historical patient's mucosal flap operation parameters, structure feature parameters, preoperative physiological data and blood vessel perfusion feature parameters, so as to obtain the predicted blood vessel perfusion feature parameters of the current patient, and then plans the blood vessel pedicle path in combination with the change characteristics of each candidate blood vessel pedicle path of the current patient. The technical problem of poor preoperative planning effect of the nasal septum mucosal flap blood vessel pedicle path is solved, the physiological structure, intraoperative operation and blood vessel dynamic perfusion information of the patient can be associated in multiple dimensions, the flap body can survive while ensuring that the path structure is reasonable, and the preoperative planning effect of the nasal septum mucosal flap blood vessel pedicle path is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of surgical path planning, in particular to a preoperative intelligent planning method for a nasal septum mucosa flap blood vessel pedicle path. BACKGROUND

[0002] The nasal septum mucosa flap is a commonly used pedicle local mucosa flap in current skull base endoscopic surgery, which is mainly supplied with blood by the sphenopalatine artery and its posterior nasal septum branch. Related personnel need to ensure the complete protection of the posterior nasal septum artery blood supply path when extracting the nasal septum mucosa flap, and the preoperative understanding of the blood vessel pedicle path helps to reduce accidental intraoperative injury, so as to ensure that the extracted nasal septum mucosa flap has sufficient blood supply, determine the length of the extractable nasal septum mucosa flap, and select appropriate repair strategies for the distal defect. Therefore, accurate preoperative planning of the blood vessel pedicle path is of great significance.

[0003] At present, imaging, intraoperative blood flow imaging, blood vessel segmentation based on deep learning, and three-dimensional reconstruction technology are proposed to support preoperative planning, but there are still limitations in actual application, resulting in poor preoperative planning effect of the nasal septum mucosa flap blood vessel pedicle path. For example, the preoperative planning scheme taking CTA (computed tomography angiography) and MRA (magnetic resonance angiography) as the core only plans the path according to the structural characteristics such as the presence or absence of blood vessels, thickness, etc., and cannot predict the flap pulling, turning over caused by the intraoperative blood vessel shape distortion, or the distal perfusion attenuation caused by the patient's blood pressure fluctuation, which may lead to the structure of the flap being complete but the function being invalid. At the same time, it also ignores the differences in the patient's basic hemodynamics, and uses a unified planning standard, resulting in significant differences in prognosis of the same path in different patients. SUMMARY

[0004] In order to solve the technical problem of poor preoperative planning effect of the nasal septum mucosa flap blood vessel pedicle path, the purpose of the present application is to provide a preoperative intelligent planning method for a nasal septum mucosa flap blood vessel pedicle path, and the technical solution adopted is as follows:

[0005] A preoperative intelligent planning method for a nasal septum mucosa flap blood vessel pedicle path, the method comprising:

[0006] Obtaining the head and neck blood vessel tree model, the nasal septum mucosa flap model and the preoperative physiological data of the current patient and the historical patients, and obtaining the structural characteristic parameters based on the geometric characteristics of the head and neck blood vessel tree model and the nasal septum mucosa flap model; obtaining the flap operation parameters and the blood vessel perfusion characteristic parameters in the operation of each historical patient;

[0007] Between the current patient and each historical patient, the reference weight of each historical patient is determined according to the differences between the head and neck blood vessel tree models and the differences between the structural characteristic parameters;

[0008] training a perfusion prediction model based on the membrane flap operation parameters, the structure feature parameters, the preoperative physiological data and the blood vessel perfusion feature parameters of all historical patients; inputting the artificial planning membrane flap operation parameters, the structure feature parameters and the preoperative physiological data of the current patient into the trained perfusion prediction model, and obtaining the predicted blood vessel perfusion feature parameters of the current patient in combination with the reference weight of the historical patients;

[0009] In the head and neck blood vessel tree model of the current patient, all candidate blood vessel pedicle paths are determined, and a planning blood vessel pedicle path is determined from all candidate blood vessel pedicle paths according to the predicted blood vessel perfusion feature parameters and the change characteristics of each candidate blood vessel pedicle path.

[0010] Further, the method for obtaining the blood vessel perfusion feature parameters comprises:

[0011] obtaining an ICG perfusion image sequence of the blood vessel of each historical patient during surgery; the ICG perfusion image sequence at least includes ICG perfusion images of the nasal septum mucosa flap at three operation nodes of flap body separation, pulling and turning over;

[0012] determining a flap body region in each frame of the ICG perfusion images; in the ICG perfusion image sequence, the flap body region with the maximum fluorescence intensity is taken as a target region, the image sequence number of the frame corresponding to the target region is taken as a peak time, a perfusion speed is determined based on the fluorescence intensity of the target region and the peak time, and a perfusion uniformity parameter is determined based on the discrete characteristics of the fluorescence intensity of the target region; a vector composed of the peak time, the perfusion speed and the perfusion uniformity parameter is taken as a perfusion feature parameter.

[0013] Further, the method for obtaining the structure feature parameters based on the geometric characteristics of the head and neck blood vessel tree model and the nasal septum mucosa flap model comprises:

[0014] obtaining blood vessel feature parameters in the head and neck blood vessel tree model and flap body feature parameters in the nasal septum mucosa flap model; the blood vessel feature parameters at least include trunk diameter, distal end minimum diameter, average bifurcation angle and total length of the blood vessel tree, and the flap body feature parameters at least include flap body surface area and flap body thickness;

[0015] a vector composed of the blood vessel feature parameters and the flap body feature parameters is taken as a structure feature parameter.

[0016] Further, the method for determining the reference weight of each historical patient according to the differences between the head and neck blood vessel tree models and the differences between the structure feature parameters comprises:

[0017] for each of the head and neck blood vessel tree models, a blood vessel structure complexity coefficient is obtained according to the diameter change of the blood vessel segments corresponding to adjacent bifurcation points and the bifurcation characteristics of the blood vessels in the head and neck blood vessel tree model;

[0018] The tree edit distance between the corresponding head and neck vascular tree models between the current patient and each historical patient is taken as a first difference parameter, the difference between the corresponding structure feature parameters is taken as a second difference parameter, and the difference between the corresponding vascular structure complexity coefficients is taken as a third difference parameter; and the first difference parameter, the second difference parameter and the third difference parameter are fused to obtain a reference weight of each historical patient relative to the current patient.

[0019] Further, for each head and neck vascular tree model, a vascular structure complexity coefficient is obtained according to the diameter change of the corresponding blood vessel segment of the adjacent bifurcation points and the bifurcation characteristics of the blood vessel.

[0020] Based on the blood flow direction, the corresponding blood vessel segment of the adjacent bifurcation points in each head and neck vascular tree model is obtained, and the bifurcation deformation parameter of each blood vessel segment is obtained according to the diameter change and bending characteristics of each blood vessel segment and the bifurcation angle at the bifurcation point corresponding to the starting point of the blood vessel segment.

[0021] In the topological structure of each head and neck vascular tree model, the vascular structure complexity coefficient is determined according to the difference in the direction of each blood vessel segment and the remaining blood vessel segments in the same topological level and the bifurcation deformation parameter, and the difference in the number of branches corresponding to different nodes in each topological level.

[0022] Further, the bifurcation deformation parameter of each blood vessel segment is obtained according to the diameter change and bending characteristics of each blood vessel segment and the bifurcation angle at the bifurcation point corresponding to the starting point of the blood vessel segment.

[0023] Each blood vessel segment is divided into a preset number of sub-segments and sorted by label.

[0024] In each blood vessel segment, the diameter mutation parameter of the blood vessel segment is obtained according to the difference between the diameters corresponding to each sub-segment and the adjacent next sub-segment.

[0025] The cosine value of the minimum bifurcation angle at the bifurcation point corresponding to the starting point of each blood vessel segment is taken as the bifurcation influence parameter.

[0026] The bifurcation deformation parameter is obtained by fusing the curvature of the blood vessel segment, the diameter mutation parameter and the bifurcation influence parameter.

[0027] Further, the vascular structure complexity coefficient is determined according to the difference in the direction of each blood vessel segment and the remaining blood vessel segments in the same topological level and the bifurcation deformation parameter, and the difference in the number of branches corresponding to different nodes in each topological level.

[0028] The topological structure of the head and neck vascular tree model is constructed with the blood vessel bifurcation point as the node and the blood vessel segment as the connection path, and the direction vector of each connection path corresponding to the blood vessel segment is obtained based on the spatial position of the node.

[0029] Taking any blood vessel segment as a target blood vessel segment, taking the mean of the sine values of the included angles between the target blood vessel segment and the corresponding strike vectors of each of the remaining blood vessel segments in the same topological level as the shape deviation parameter of the target blood vessel segment;

[0030] Fusing the bifurcation deformation parameter and the shape deviation parameter to obtain a structure complexity parameter of the target blood vessel segment;

[0031] Taking the coefficient of variation of the total number of the connection paths of all nodes in each topological level as a branch disorder parameter of each topological level;

[0032] Weighting the structure complexity parameters of all blood vessel segments in each topological level by using the branch disorder parameter of the topological level, and taking the normalized value of the weighted sum result as a blood vessel structure complexity coefficient of the head and neck blood vessel tree model.

[0033] Further, the inputting of the artificial planning flap operation parameter of the current patient, the structure feature parameter and the preoperative physiological data into the trained perfusion prediction model, combined with the reference weight of the historical patient, to obtain the predicted blood vessel perfusion feature parameter of the current patient, comprises:

[0034] Taking the flap operation parameter, the structure feature parameter and the preoperative physiological data of all historical patients as feature data, and taking the corresponding blood vessel perfusion feature parameter as the label of the feature data, constructing sample data to train the perfusion prediction model;

[0035] Inputting the artificial planning flap operation parameter of the current patient, the structure feature parameter and the preoperative physiological data into the trained perfusion prediction model to obtain the blood vessel perfusion feature parameter of the current patient;

[0036] Based on the reference weight, a reference patient is selected from all historical patients, and the feature data of the reference patient is input into the trained perfusion prediction model to obtain the blood vessel perfusion feature parameter of the reference patient;

[0037] The difference between the blood vessel perfusion feature parameter of the current patient and the blood vessel perfusion feature parameter of the reference patient is calculated, and the difference is weighted by using the reference weight of the reference patient and then averaged to obtain the output deviation of the perfusion prediction model;

[0038] The sum of the blood vessel perfusion feature parameter of the current patient and the corresponding output deviation is taken as the predicted blood vessel perfusion feature parameter of the current patient.

[0039] Further, the determining of the planning blood vessel pedicle path from all candidate blood vessel pedicle paths according to the predicted blood vessel perfusion feature parameter and the change feature of each candidate blood vessel pedicle path comprises:

[0040] For the current patient, obtain the artificial flap operation parameters corresponding to each candidate blood vessel pedicle path, and input the artificial flap operation parameters, the structural characteristic parameters and the preoperative physiological data into the trained perfusion prediction model to obtain the blood vessel perfusion characteristic parameters output by the model;

[0041] For each candidate blood vessel pedicle path, obtain the perfusion safety parameters according to the deviation of the blood vessel perfusion characteristic parameters relative to the predicted blood vessel perfusion characteristic parameters;

[0042] According to the total length of the candidate blood vessel pedicle path and the structural complexity parameters of each blood vessel segment on the path, obtain the path complexity parameters;

[0043] According to the perfusion safety parameters and the path complexity parameters, obtain the path candidate score;

[0044] Based on the path candidate score, filter out the planned blood vessel pedicle path from all the candidate blood vessel pedicle paths.

[0045] Further, the flap operation parameters at least include the pulling length and the turning angle of the nasal septum mucosal flap during each intraoperative operation.

[0046] The present application has the following beneficial effects:

[0047] The application firstly acquires the head and neck vascular tree model, the nasal septum mucosa flap model, the preoperative physiological data and the structural feature parameters of the current patient and each historical patient, simultaneously acquires the intraoperative mucosa flap operation parameter and the blood vessel perfusion feature parameter of each historical patient, and prepares for subsequent analysis of the blood flow information of the blood vessel structure and the mucosa flap operation process of the current patient to plan the blood vessel pedicle path; then, the reference weight of each historical patient is acquired based on the physiological structure similarity, a historical similar reference is provided for the current patient; further, the perfusion prediction model is trained based on the mucosa flap operation parameter, the structural feature parameter, the preoperative physiological data and the blood vessel perfusion feature parameter of all historical patients, then the artificial planning mucosa flap operation parameter, the structural feature parameter and the preoperative physiological data of the current patient are input into the trained perfusion prediction model, and the prediction deviation of the perfusion prediction model is corrected by combining the reference weight of the historical patient, so that the predicted blood vessel perfusion feature parameter of the current patient is accurately acquired; finally, all candidate blood vessel pedicle paths are determined in the head and neck vascular tree model of the current patient, and the perfusion safety and the path complexity are evaluated according to the predicted blood vessel perfusion feature parameter and the change characteristics of each candidate blood vessel pedicle path, so that the planning blood vessel pedicle path is determined from all candidate blood vessel pedicle paths. The physiological structure, the intraoperative operation and the blood vessel dynamic perfusion information of a large number of historical patients are associated in multiple dimensions, and the prediction result of the perfusion prediction model is combined as a constraint condition to plan the blood vessel pedicle path, which breaks through the planning logic of the traditional scheme according to the blood vessel structure only, can reduce the probability of the path structure being reasonable but the function being unsafe, ensures the flap to survive while ensuring the path structure to be reasonable, and improves the preoperative planning effect of the nasal septum mucosa flap blood vessel pedicle path. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creating laborious work.

[0049] Figure 1 The flowchart of a preoperative intelligent planning method of a nasal septum mucosa flap blood vessel pedicle path provided by the embodiments of the present application;

[0050] Figure 2 The flowchart of a reference weight acquisition method provided by the embodiments of the present application;

[0051] Figure 3 The flowchart of a blood vessel structure complexity coefficient acquisition method provided by the embodiments of the present application. DETAILED DESCRIPTION

[0052] In order to further illustrate the technical means and effects taken by the present application to achieve the predetermined object, the following describes in detail the specific implementation, structure, features and effects of the preoperative intelligent planning method for a nasal septum mucosa flap blood vessel pedicle path according to the present application, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0054] The following describes a specific scheme of the preoperative intelligent planning method for a nasal septum mucosa flap blood vessel pedicle path according to the present application in combination with the accompanying drawings.

[0055] Please refer to Figure 1 which shows a flowchart of the preoperative intelligent planning method for a nasal septum mucosa flap blood vessel pedicle path, including steps S1 to S4, as follows.

[0056] Step S1, obtain the head and neck blood vessel tree model, nasal septum mucosa flap model and preoperative physiological data of the current patient and historical patients, and obtain the structural feature parameters based on the geometric features of the head and neck blood vessel tree model and the nasal septum mucosa flap model. Obtain the flap operation parameters and blood vessel perfusion feature parameters of each historical patient in the operation.

[0057] In order to reasonably plan the nasal septum mucosa flap blood vessel pedicle path before operation to ensure the survival of the nasal septum mucosa flap, in one embodiment of the present application, the head and neck blood vessel tree model, nasal septum mucosa flap model and preoperative physiological data of the current patient and each historical patient are first obtained, and the flap operation parameters and blood vessel perfusion feature parameters of each historical patient in the historical operation are also obtained, to prepare for subsequent analysis of the blood vessel structure to determine the blood vessel pedicle path.

[0058] It should be noted that the above data acquisition process for each patient is consistent, and only one patient is taken as an example for analysis and description here, and will not be described one by one.

[0059] Specifically, the CTA image and MRI image of the patient at the head and neck site before operation are obtained from the electronic medical record database, and the image is required to cover at least the whole nasal septum and the sphenopalatine foramen blood vessel pedicle origin area, and the DICOM format image sequence is output respectively. Then each frame of image is processed for denoising, artifact elimination and the like, to ensure that the blood vessel edge details are retained while the speckle noise is suppressed, and the small blood vessel structure can be additionally enhanced by applying Frangi blood vessel filtering, so as to output the enhanced CTA image and MRI image.

[0060] The blood vessels and nasal septum mucosal flaps in each enhanced CTA image are extracted based on a deep learning model such as U-Net++, and the blood vessels in the two-dimensional slices (enhanced CTA images) are converted into a three-dimensional model by using a medical image processing software (such as 3D Slicer, an open-source medical image processing software), and then the center line of the blood vessels in the three-dimensional model is extracted by using a skeleton extraction algorithm such as Zhang-Suen thinning method (a classical skeleton extraction algorithm), the bifurcation points of the blood vessel structure are marked, and a (three-dimensional) head and neck blood vessel tree model containing the diameter, bifurcation angle and curvature is generated, which is prepared for subsequent analysis of hemodynamics. In order to improve the analysis efficiency, the implementer can also directly take the sphenopalatine foramen as the starting bifurcation point to construct a blood vessel bundle model, thereby reducing unnecessary subsequent blood vessel analysis.

[0061] The nasal septum mucosal flap area in each enhanced MRI image is segmented based on a deep learning model such as U-Net++, and the nasal septum mucosal flap area in the enhanced CTA image is combined to reconstruct a (three-dimensional) nasal septum mucosal flap model. The survival of the flap (nasal septum mucosal flap, hereinafter referred to as flap) is related to the blood supply capacity, which provides a reference basis for subsequent planning of the blood vessel pedicle path.

[0062] It should be noted that the acquisition and preprocessing of CTA images and MRI images, the construction of head and neck blood vessel tree model and nasal septum mucosal flap model are all prior art means, and the specific process will not be described.

[0063] After obtaining the head and neck blood vessel tree model and the nasal septum mucosal flap model of the patient, the geometric feature parameters can be further obtained based on the geometric features of the two models. The structural feature parameters reflect the structural complexity of the patient's head and neck blood vessels and the flap, which prepares for subsequent planning of the blood vessel pedicle path.

[0064] Preferably, in an embodiment of the present application, the method for obtaining the structural feature parameters comprises: obtaining the blood vessel feature parameters in the head and neck blood vessel tree model and the flap feature parameters in the nasal septum mucosal flap model. The blood vessel feature parameters at least include the main stem diameter, the minimum distal diameter, the average bifurcation angle and the total length of the blood vessel tree, and the flap feature parameters at least include the flap surface area and the flap thickness. The vector composed of the blood vessel feature parameters and the flap feature parameters is taken as the structural feature parameters.

[0065] It should be noted that the acquisition of the above parameters and the construction of the vector are all prior art means, and will not be described in detail. In other embodiments, an array can be used instead of a vector to represent the structural feature parameters.

[0066] Meanwhile, the monitoring data of the patient within 24 hours before the surgery under the preoperative physiological indicators is extracted from the electronic medical record, the monitoring data is cleaned (including removal and processing of abnormal values, missing values and repeated data), and preoperative physiological data is obtained. The preoperative physiological indicators at least include blood pressure, heart rate, and vascular elasticity index.

[0067] It should be noted that the collection and cleaning operation of the preoperative physiological data is a prior art means, which will not be repeated, and the implementer can also add other preoperative physiological indicators or adjust the preoperative monitoring period.

[0068] Considering that the survival of the nasal septum mucosal flap depends on the integrity of the vascular structure, the effectiveness of hemodynamics, and the individual physiological tolerance, although the head and neck vascular tree model can provide anatomical structure information of the blood vessels, it cannot provide blood flow state information. If the blood circulation in the blood vessels is blocked, it will still affect the survival of the flap. Considering that the flap will be pulled or turned to a certain extent during the operation, at this time the blood flow state will be affected, which may affect the survival rate of the flap. Therefore, one embodiment of the present application further acquires the mucosal flap operation parameters of each historical patient during the operation and the vascular perfusion feature parameters, wherein the mucosal flap operation parameters represent the pulling and turning of the flap during the operation, so as to evaluate the influence of the intraoperative operation on the blood flow in combination with the blood flow perfusion feature parameters, and the vascular perfusion feature parameters represent the blood flow state information, which provides a historical reference for subsequent evaluation and planning of the vascular pedicle path of the current patient.

[0069] It should be noted that the above historical patient is a postoperative flap survival patient.

[0070] Specifically, the flap operation record is obtained from the operation record of each historical patient, and the operation parameters and time corresponding to each operation step are obtained from the flap operation record. The operation steps include the separation, pulling and turning of the flap by the operator during the operation. Therefore, the mucosal flap operation parameters at least include the separation, pulling length and turning angle of the nasal septum mucosal flap during each intraoperative operation.

[0071] It should be noted that in other embodiments, the implementer can also sort each operation step and the corresponding operation parameters based on the time sequence order to obtain a mucosal flap operation parameter sequence, which can also be regarded as a mucosal flap operation parameter. The acquisition of the mucosal flap operation parameter is a prior art means, which will not be repeated.

[0072] Preferably, in one embodiment of the present application, ICG (indocyanine green) is considered as a tracer, which can flow with blood circulation during the operation, so as to characterize the blood flow state. In addition, the time of ICG flowing from the main blood vessel to the valve body area reflects the blood flow rate, and the slow blood flow rate can cause valve body necrosis, and uneven blood flow distribution can also cause partial valve body necrosis. Therefore, by analyzing the uniformity of the fluorescence intensity in the valve body area and the time of reaching the maximum value, the blood flow rate and the uniform distribution of blood flow can be evaluated.

[0073] The method for obtaining the vascular perfusion feature parameter comprises: obtaining an ICG perfusion image sequence of each historical patient's vascular during operation. The ICG perfusion image sequence at least includes ICG perfusion images of the nasal septum mucosa valve at three operation nodes of valve body separation, pulling and turning. The valve body area in each frame of the ICG perfusion image is determined. In the ICG perfusion image sequence, the valve body area with the maximum fluorescence intensity is taken as a target area, the frame number corresponding to the target area is taken as a peak time, the perfusion rate is determined based on the fluorescence intensity of the target area and the peak time, and the perfusion uniformity parameter is determined based on the dispersion characteristics of the fluorescence intensity of the target area. The vector composed of the peak time, the perfusion rate and the perfusion uniformity parameter is taken as the perfusion feature parameter.

[0074] Specifically, the ICG perfusion video of each historical patient's vascular during operation is obtained from the operation record of the historical patient, and the ICG perfusion video is subjected to frame processing to obtain an ICG perfusion image sequence. The ICG perfusion image sequence at least includes ICG perfusion images of the nasal septum mucosa valve at three operation nodes of valve body separation, pulling and turning, i.e. at least includes blood flow information in the operation process of the mucosa valve, thereby providing a reference for subsequent evaluation of the planned path in combination with the planned mucosa valve operation of the current patient.

[0075] The valve body area in each frame of the ICG perfusion image is extracted based on a deep learning model such as U-Net++ (an improved semantic segmentation neural network), and the mean value of the fluorescence intensity of all pixel points in the valve body area is taken as the fluorescence intensity of the valve body area. In the ICG perfusion image sequence, the valve body area with the maximum fluorescence intensity is taken as a target area, and the frame number corresponding to the target area is taken as a peak time. The fluorescence intensity of the target area is divided by the peak time to obtain a perfusion rate. The negative correlation mapping result of the coefficient of variation of the fluorescence intensity of the pixel points in the target area is added to a minimum positive parameter 0.001 to obtain an inverse number, thereby obtaining a perfusion uniformity parameter. The peak time, the perfusion rate and the perfusion uniformity parameter are taken as vector elements, and the vector composed of the vector elements is taken as the perfusion feature parameter. The perfusion feature parameter reflects the blood flow state in the operation process of the nasal septum mucosa valve of the historical patient.

[0076] It should be noted that the ICG perfusion and image acquisition and processing, and the extraction of the flap region are existing technical means, and will not be described in detail. In other embodiments, the implementer can also use other perfusion imaging techniques to obtain blood flow state information, and can also use other negative correlation mapping methods instead of the reciprocal operation.

[0077] Step S2, between the current patient and each historical patient, according to the differences between the head and neck vascular tree models and the differences between the structure characteristic parameters, the reference weight of each historical patient is determined.

[0078] Considering the flap operation parameters and perfusion characteristic parameters of each historical patient during the septal mucosal flap surgery, a certain planning reference can be provided for the surgical path of the current patient, that is, under what kind of flap operation can ensure that the blood flow flows to the flap in time, so as to ensure the survival of the flap. When the blood vessel structure and the flap structure of the historical patient and the current patient are similar, the planning reference value provided by the historical patient is also large, so that the perfusion characteristic parameters of the current patient can be predicted in combination with the physiological structure, the preoperative physiological index and the perfusion characteristic parameters of the historical patient, and then a safe reference can be provided for the path planning of the current patient to avoid unreasonable planning path leading to flap necrosis.

[0079] Based on this, the reference weight of each historical patient is obtained in the embodiments of the present application, which preliminarily reflects the similarity between the blood vessel structure and the flap structure of the current patient and the historical patient, and reflects the reference value provided by the historical patient, preparing for subsequent path planning.

[0080] Preferably, in one embodiment of the present application, the method for obtaining the reference weight is as shown in Figure 2 , including steps S201 to S202, which are as follows.

[0081] Step S201, for each head and neck vascular tree model, according to the diameter change of the adjacent bifurcation point corresponding blood vessel segment and the bifurcation characteristics of the blood vessel, the blood vessel structure complexity coefficient is obtained.

[0082] Considering that the blood vessel structure determines the basic potential of blood perfusion, when the blood vessel structure is complex, characteristics such as high branch density, abnormal bifurcation angle, sudden change of pipe diameter, excessive bending, etc. will increase the resistance of blood transmission, reduce the blood flow speed in the blood vessel network and the uniformity of blood flow distribution, and thus may lead to flap necrosis. Based on this, one embodiment of the present application first obtains the structure complexity parameter of the blood vessel, which reflects the blood perfusion ability and provides a certain reference value for subsequent screening of historical reference patients with similar blood perfusion ability.

[0083] In a preferred embodiment of the present application, in order to analyze the diameter change characteristics, bifurcation characteristics and bending characteristics of the vascular tree, each vascular segment can be determined first, and then the bifurcation deformation parameters of each vascular segment are evaluated. Considering that the topological connection of the blood vessels can also reflect the functional redundancy of the vascular network, the ordered topological structure is conducive to blood flow distribution, and the disordered topological structure is easy to reduce the perfusion efficiency, the topological structure of the vascular tree can be further analyzed, so as to comprehensively evaluate the structure complexity parameters of the vascular tree in combination with the bifurcation deformation parameters of the vascular segment.

[0084] Referring to Figure 3 which shows a flowchart of a method for obtaining a vascular structure complexity coefficient, specifically comprising steps S211 to S212, as follows.

[0085] In step S211, the blood vessel segments corresponding to adjacent bifurcation points in each head and neck vascular tree model are obtained based on the blood flow direction, the diameter change and bending characteristics of each blood vessel segment, and the bifurcation angle at the bifurcation point corresponding to the starting point of the blood vessel segment, and the bifurcation deformation parameters of the blood vessel segment are obtained.

[0086] Taking any head and neck vascular tree model as an example for analysis and description, first, along the blood flow direction, the corresponding segment of each bifurcation point in the head and neck vascular tree model to the next bifurcation point is taken as a blood vessel segment.

[0087] In order to analyze the diameter mutation characteristics of each blood vessel segment, each blood vessel segment is first divided into a plurality of sub-segments, and then the diameter mutation characteristics can be evaluated according to the diameter change between adjacent sub-segments to obtain the diameter mutation parameters. Considering that the smaller the blood vessel bifurcation angle corresponding to the branch where the blood vessel segment is located, the more likely it is to cause vortex flow at the branch, thereby causing uneven blood flow distribution. Also considering that the greater the curvature of the blood vessel segment, the more likely it is to affect the blood flow velocity. Based on this, in a preferred embodiment of the present application, the method for obtaining the bifurcation deformation parameters comprises: dividing each blood vessel segment into a preset number of sub-segments and sorting the labels. In each blood vessel segment, the diameter mutation parameters of the blood vessel segment are obtained according to the difference between the diameters corresponding to each sub-segment and the adjacent next sub-segment. The cosine value of the smallest bifurcation angle at the bifurcation point corresponding to the starting point of each blood vessel segment is taken as the bifurcation influence parameter. The curvature, diameter mutation parameters and bifurcation influence parameters of the blood vessel segment are fused to obtain the bifurcation deformation parameters.

[0088] Specifically, the preset number is set to 10, and the implementer can also adjust it. Starting from the bifurcation starting point corresponding to the blood vessel segment, the sub-segments are sorted and labeled, and the sub-segment number is larger and larger along the blood flow direction. The diameter of the middle segment in each sub-segment is taken as the sub-segment diameter, the difference between the diameters corresponding to each sub-segment and the adjacent next sub-segment is calculated, and the difference is divided by the diameter of each sub-segment to obtain the diameter change rate of each sub-segment. The maximum diameter change rate is taken as the diameter mutation parameter of the blood vessel segment.

[0089] Then, the cosine value of the minimum bifurcation angle at the bifurcation point corresponding to the start point of each blood vessel segment is taken as the bifurcation influence parameter. The smaller the minimum bifurcation angle is, the larger the cosine value is, and the greater the influence of the bifurcation on the blood flow is, and the more likely the blood flow distribution is uneven. Then, the curvature of each blood vessel segment in the head and neck blood vessel tree model is determined, the curvature is linearly normalized, and the normalized value, the diameter mutation parameter, and the bifurcation influence parameter are multiplied and fused to obtain the bifurcation deformation parameter of the blood vessel segment. It should be noted that linear normalization is a known technology and will not be described here.

[0090] In step S212, in the topological structure of each head and neck blood vessel tree model, the blood vessel structure complexity coefficient is determined according to the orientation difference between each blood vessel segment and the remaining blood vessel segments in the same topological level and the bifurcation deformation parameter, and the difference in the number of branches corresponding to different nodes in each topological level.

[0091] It is considered that the greater the difference in the number of blood vessel branches at different bifurcation nodes of each level in the topological structure of the head and neck blood vessel tree model is, the stronger the topological structure disorder of the blood vessel is, and the greater the probability of uneven distribution of the blood transmission process is. At the same time, the greater the orientation difference of each level of different blood vessel branches and the greater the bifurcation deformation parameter of each blood vessel branch are, the more abnormal the structure and orientation of the blood vessel branch are, and the more complex and disordered the topological structure is. Based on this, the blood vessel structure complexity coefficient can be comprehensively evaluated.

[0092] In a preferred embodiment of the present application, the blood vessel structure complexity coefficient is determined according to the orientation difference between each blood vessel segment and the remaining blood vessel segments in the same topological level and the bifurcation deformation parameter, and the difference in the number of branches corresponding to different nodes in each topological level, including: taking the bifurcation point of the blood vessel as the node and the blood vessel segment as the connection path to construct the topological structure of the head and neck blood vessel tree model, and obtaining the orientation vector of the blood vessel segment corresponding to each connection path based on the spatial position of the node. Taking any blood vessel segment as a target blood vessel segment, the mean value of the sine value of the included angle between the orientation vector corresponding to the target blood vessel segment and the remaining blood vessel segments in the same topological level is taken as the orientation deviation parameter of the target blood vessel segment. The bifurcation deformation parameter and the orientation deviation parameter are fused to obtain the structure complexity parameter of the target blood vessel segment. The variation coefficient of the total number of connection paths corresponding to all nodes in each topological level is taken as the branch disorder parameter of each topological level. The structure complexity parameters of all blood vessel segments in each topological level are weighted by the branch disorder parameter of the topological level, and the normalized value of the weighted sum result is taken as the blood vessel structure complexity coefficient of the head and neck blood vessel tree model.

[0093] Specifically, first, the topology of the head and neck vascular tree model is constructed, and the heading vector of each connected path corresponding to the vascular segment in the topology is calculated, which is a known technology and will not be described again. Among them, the direction of the heading vector is consistent with the direction of blood flow. Taking any vascular segment (connected path) as a target vascular segment, and taking the structural complexity parameter of the target vascular segment as an example for analysis and description.

[0094] The topology level where the target vascular segment is located is determined, and the mean value of the sine value of the angle between the target vascular segment and the heading vector corresponding to each vascular segment in the same topology level is taken as the shape deviation parameter of the target vascular segment. The greater the angle, the greater the difference in the heading of the blood vessel, and the more complex and disordered the topology structure of the level is, and the greater the shape deviation parameter is. Then, the bifurcation deformation parameter and the shape deviation parameter of the target vascular segment are multiplied to obtain the structural complexity parameter. The structural complexity parameter comprehensively considers the bifurcation deformation and the heading abnormality of each vascular segment, and reflects the local structural complexity of the blood vessel.

[0095] Then, in each topology level, the subordinate connection paths (vascular segments starting from the bifurcation point) of each node (bifurcation point) are determined, the total number of subordinate connection paths of each node is counted, and then the coefficient of variation of the total number of subordinate connection paths corresponding to all nodes in each topology level is calculated to obtain the branch disorder parameter of each topology level. The greater the coefficient of variation, the greater the difference in the number of blood vessel branches at different bifurcation points in the topology level, and the more complex and disordered the topology structure of the level is, and the greater the branch disorder parameter is. Then, the structural complexity parameter of each vascular segment in the level is weighted using the branch disorder parameter of each topology level, and the weighted sum result is linearly normalized to obtain the vascular structure complexity coefficient of the head and neck vascular tree model.

[0096] It should be noted that the linear normalization is performed in the dimension of all patients, and the acquisition of the coefficient of variation is a known technology and will not be described again.

[0097] Step S202, between the current patient and each historical patient, the tree edit distance between the corresponding head and neck vascular tree models is taken as the first difference parameter, the difference between the corresponding structural feature parameters is taken as the second difference parameter, and the difference between the corresponding vascular structure complexity coefficients is taken as the third difference parameter. The reference weight of each historical patient relative to the current patient is obtained by fusing the first difference parameter, the second difference parameter and the third difference parameter.

[0098] It is considered that the tree edit distance between the head and neck vascular tree model between the current patient and each historical patient to some extent reflects the similarity between the overall vascular structures, the difference between the vascular structure complexity coefficients further comprehensively reflects the vascular structure similarity from the local details of the blood vessels, and the difference between the structure characteristic parameters comprehensively reflects the similarity between the vascular structure characteristics and the valve structure characteristics. The above three can be integrated to evaluate the structural similarity between the patients, and then evaluate the physiological structure reference value of the historical patients to the current patient, and determine the reference weight of each historical patient.

[0099] Specifically, the acquisition of the tree edit distance is a known technology and will not be described again, so as to obtain the first difference parameter. Since the structure characteristic parameters are in the form of a vector, the cosine distance can be directly used as the second difference parameter. The absolute value of the difference is used to measure the difference between the vascular structure complexity coefficients, and the third difference parameter is obtained. The three difference parameters are multiplied and fused, the product is linearly normalized, and then the normalized value is subtracted by a constant 1 to adjust the logical relationship, and the reference weight of each historical patient is obtained. When the product is smaller, the difference between the historical patient and the reference patient is smaller, and the reference weight of the historical patient is larger.

[0100] It should be noted that the cosine distance and linear normalization are both prior art and will not be described again.

[0101] Step S3, based on the membrane flap operation parameters, structure characteristic parameters, preoperative physiological data and vascular perfusion characteristic parameters of all historical patients, a perfusion prediction model is trained. The artificial planning membrane flap operation parameters, structure characteristic parameters and preoperative physiological data of the current patient are input into the trained perfusion prediction model, and the reference weight of the historical patient is combined to obtain the predicted vascular perfusion characteristic parameters of the current patient.

[0102] It is considered that the structure characteristic parameters and preoperative physiological data of the historical patient comprehensively reflect the physiological condition, the membrane flap operation parameters reflect the operation of the valve body during the operation, and the influence information of the blood flow (surgical traction reversal causes the deformation of the blood vessels, thereby affecting the blood flow) is provided. The vascular perfusion characteristic parameters directly reflect the specific information of the blood flow to the valve area under the corresponding intraoperative operation.

[0103] Exemplarily, the membrane flap operation parameters at least include the traction length and the turning angle of the nasal septum mucosa flap at each intraoperative operation.

[0104] The neural network model can learn the basic correlation between the physiological structure of a large number of historical patients, intraoperative operation and blood flow dynamics, and then predict the blood flow dynamics of the current patient based on the physiological structure of the current patient and the intraoperative operation evaluated by the medical staff, thereby avoiding valve necrosis. The more similar the physiological structure of the historical patient to the current patient, the greater the reference value of the basic correlation provided by the historical patient, and thus the predicted vascular perfusion characteristic parameter of the current patient can be accurately obtained.

[0105] Preferably, in an embodiment of the present application, the model is first trained to predict the vascular perfusion parameter of the current patient. It is also considered that the model training learns the physiological structure and vascular perfusion characteristics of a large number of historical patients with different physiological structures, and the obtained average optimal correlation may result in a certain systematic deviation in the prediction result. The vascular perfusion characteristic reference value provided by the historical patient with similar physiological structure is greater, thereby improving the accuracy of the prediction result. Directly using similar historical reference patients to train the model can make the prediction result more accurate, but the model needs to be retrained for the evaluation of each current patient, which reduces the prediction efficiency. Based on this, the method for obtaining the vascular perfusion characteristic parameter includes: taking the leaflet operation parameter, the structure characteristic parameter and the preoperative physiological data of all historical patients as feature data, taking the corresponding vascular perfusion characteristic parameter as the label of the feature data, constructing sample data to train the perfusion prediction model. The artificial planning leaflet operation parameter, the structure characteristic parameter and the preoperative physiological data of the current patient are input into the trained perfusion prediction model to obtain the vascular perfusion characteristic parameter of the current patient. The reference weight is used to select the reference patient from all historical patients, and the feature data of the reference patient is input into the trained perfusion prediction model to obtain the vascular perfusion characteristic parameter of the reference patient. The difference between the vascular perfusion characteristic parameter of the current patient and the vascular perfusion characteristic parameter of the reference patient is calculated, and the difference is weighted by the reference weight of the reference patient and then averaged to obtain the output deviation of the perfusion prediction model. The sum of the vascular perfusion characteristic parameter of the current patient and the corresponding output deviation is taken as the predicted vascular perfusion characteristic parameter of the current patient.

[0106] Specifically, the leaflet operation parameter, the structure characteristic parameter and the preoperative physiological data of each historical patient are taken as a feature data, and the corresponding vascular perfusion characteristic parameter is taken as the label of the feature data. The sample data is constructed, and the sample data is divided into a training set and a validation set in a ratio of 7:3. The CNN-LSTM (combination model of convolutional neural network and long short-term memory network) model is taken as the basic architecture, the MSE (mean square error) is taken as the loss function, and the trained perfusion prediction model is obtained when the error of the validation set is less than 9%.

[0107] Then the membrane flap operation plan of the relevant medical staff is obtained, and the artificial planning membrane flap operation parameter is determined, which is a known technology and will not be described here. The artificial planning membrane flap operation parameter, the structural feature parameter and the preoperative physiological data of the current patient are input into the trained perfusion prediction model to obtain the blood vessel perfusion feature parameter output by the model. The blood vessel perfusion feature parameter is the average prediction result directly output by the model, and there may be a certain deviation.

[0108] The threshold is further used to screen out historical patients with larger reference weights, for example, the threshold is set to 0.6, and the implementer can also adjust it. The historical patients with reference weights greater than or equal to 0.6 are taken as reference patients. For each reference patient, the feature data (membrane flap operation parameter, structural feature parameter and preoperative physiological data) thereof is input into the trained perfusion prediction model, and then the difference between the blood vessel perfusion feature parameter output by the model and the blood vessel perfusion feature parameter of the reference patient is calculated (wherein since the blood vessel perfusion feature parameter is in the form of a vector, the cosine distance can be directly calculated to measure the difference), the cosine distance is weighted by using the reference weight of the reference patient, and the weighted average result of all reference patients is taken as the model output deviation, wherein the model output deviation can also take negative value or 0.

[0109] Finally, the blood vessel perfusion feature parameter is added to the model output deviation to obtain the predicted blood vessel perfusion feature parameter of the current patient, thereby improving the prediction accuracy.

[0110] It should be noted that the training and application of the perfusion prediction model are known technologies and will not be described here.

[0111] The predicted blood vessel perfusion feature parameter is the blood flow information of the current patient evaluated by means of historical patient reference. The predicted blood vessel perfusion feature parameter includes peak time, perfusion speed and perfusion uniformity parameter, which can be regarded as a safety threshold. Exceeding the safety threshold will lead to valve body necrosis, which prepares for subsequent screening and planning path.

[0112] Step S4, in the head and neck blood vessel tree model of the current patient, all candidate blood vessel pedicle paths are determined, and the planned blood vessel pedicle path is determined from all candidate blood vessel pedicle paths according to the predicted blood vessel perfusion feature parameter and the change feature of each candidate blood vessel pedicle path.

[0113] In the head and neck blood vessel tree model of the current patient, all candidate blood vessel pedicle paths are first generated based on the Dijkstra algorithm, wherein each candidate blood vessel pedicle path includes a blood vessel pedicle starting point (sphenopalatine foramen) and an ending point (valve body) so as to analyze and screen the planned blood vessel pedicle path subsequently. The generation of the candidate blood vessel pedicle path based on the Dijkstra algorithm is a known technology and will not be described here.

[0114] In view of the fact that not only the survival of the valve body needs to be considered during the operation, but also the difficulty of the operation and the structural complexity of the vascular pedicle path, the more complex the structure, the more difficult the operation, and unnecessary risks may be caused. Based on this, the predicted vascular perfusion characteristic parameters and the change characteristics of each candidate vascular pedicle path can be used to determine the planned vascular pedicle path from all candidate vascular pedicle paths.

[0115] Preferably, in an embodiment of the present application, considering that each candidate vascular pedicle path is known, the valve body operation can be planned to predict the vascular perfusion characteristics to evaluate the risk of valve body necrosis and obtain the perfusion safety parameter. In addition, considering that each candidate vascular pedicle path is composed of multiple vascular segments, the structural complexity parameter of the vascular segment can be combined to evaluate the path complexity, and then the path candidate score is comprehensively evaluated to screen the vascular pedicle path. Based on this, the method for obtaining the planned vascular pedicle path includes: for each candidate vascular pedicle path, obtaining the perfusion safety parameter according to the deviation of the vascular perfusion characteristic parameter relative to the predicted vascular perfusion characteristic parameter. Obtain the path complexity parameter according to the total length of the candidate vascular pedicle path and the structural complexity parameter of each vascular segment on the path. Obtain the path candidate score according to the perfusion safety parameter and the path complexity parameter. Based on the path candidate score, the planned vascular pedicle path is screened from all candidate vascular pedicle paths.

[0116] Specifically, the artificial planning valve operation parameters corresponding to each candidate vascular pedicle path can be fitted through a deep learning model or an artificial planning method. Then the artificial planning valve operation parameters, the structural characteristic parameters and the preoperative physiological data are input into the trained perfusion prediction model to obtain the path perfusion characteristic parameters output by the model. The path perfusion characteristic parameters reflect the blood flow information under the path, which prepares for the subsequent evaluation of the risk of valve body necrosis.

[0117] Then compare the path perfusion characteristic parameters of each candidate vascular pedicle path, such as (t, v, j), with the predicted vascular perfusion characteristic parameters, such as (T, V, J), to evaluate the deviation between them. For the same position number vector element, map T-t into the ReLU (Rectified Linear Unit) function, when t is greater than T, the peak time is larger, the valve body survival rate is lower, and the function value is 0. Map V-v into the ReLU function, when V is greater than v, the perfusion velocity is smaller, the valve body survival rate is lower, and the function value is 0. Map J-j into the ReLU function, when J is greater than j, the perfusion uniformity parameter is smaller, the valve body survival rate is lower, and the function value is 0. Wherein, t represents the peak time of the candidate vascular pedicle path, v represents the perfusion velocity of the candidate vascular pedicle path, j represents the perfusion uniformity parameter of the candidate vascular pedicle path, T represents the peak time in the predicted vascular perfusion characteristic parameters, V represents the perfusion velocity in the predicted vascular perfusion characteristic parameters, and J represents the perfusion uniformity parameter in the predicted vascular perfusion characteristic parameters.

[0118] The three function mapping values are accumulated and mapped into a sigmoid (logistic function) function to adjust the value range, to obtain a perfusion safety parameter. Then the total length of each candidate blood vessel pedicle path is multiplied by the cumulative sum of the structural complexity parameters of each blood vessel segment on its path, and the product is mapped into a sigmoid function to adjust the value range, to obtain a path complexity parameter.

[0119] Finally, the path complexity parameter is subtracted by a constant 1 to adjust the logic, the difference is multiplied by the perfusion safety parameter, and the product is taken as the path candidate score. The smaller the path complexity parameter, the larger the perfusion safety parameter, and the larger the path candidate score, the greater the planning possibility. The candidate blood vessel pedicle path with the largest path candidate score is taken as the planned blood vessel pedicle path.

[0120] Up to now, the preoperative intelligent planning of the nasal septum mucosa flap blood vessel pedicle path has been completed.

[0121] To sum up, the present application first obtains the head and neck blood vessel tree model, the nasal septum mucosa flap model, the preoperative physiological data and the structural feature parameters of the current patient and each historical patient, and simultaneously obtains the mucosa flap operation parameters and the blood vessel perfusion feature parameters of each historical patient in the operation, to prepare for subsequent analysis of the blood flow information of the blood vessel structure and the mucosa flap operation process of the current patient to plan the blood vessel pedicle path. Then, based on the physiological structure similarity, the reference weight of each historical patient is obtained to provide a historical similar reference for the current patient. Further, based on the mucosa flap operation parameters, the structural feature parameters, the preoperative physiological data and the blood vessel perfusion feature parameters of all historical patients, a perfusion prediction model is trained, and then the artificial planning mucosa flap operation parameters, the structural feature parameters and the preoperative physiological data of the current patient are input into the trained perfusion prediction model, and at the same time, the prediction bias of the perfusion prediction model is corrected by combining the reference weight of the historical patient, to accurately obtain the predicted blood vessel perfusion feature parameters of the current patient. Finally, in the head and neck blood vessel tree model of the current patient, all candidate blood vessel pedicle paths are determined, and according to the predicted blood vessel perfusion feature parameters and the change characteristics of each candidate blood vessel pedicle path, the perfusion safety and the path complexity are evaluated, so as to determine the planned blood vessel pedicle path from all candidate blood vessel pedicle paths. The present application correlates the physiological structure, the intraoperative operation and the blood vessel dynamic perfusion information of a large number of historical patients in multiple dimensions, and combines the prediction result of the perfusion prediction model as a constraint condition to plan the blood vessel pedicle path, which breaks through the planning logic of the traditional scheme only according to the blood vessel structure, can reduce the probability of the path structure being reasonable but the function being unsafe, ensures the survival of the flap while ensuring the rationality of the path structure, and improves the preoperative planning effect of the nasal septum mucosa flap blood vessel pedicle path.

[0122] It should be noted that the progressive order of the above-described embodiments is merely for the purpose of description, and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or can be advantageous.

[0123] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the difference from other embodiments.

Claims

1. A method for preoperative intelligent planning of nasal septum mucosal flap pedicle pathway, characterized in that, The method comprises: obtaining the head and neck blood vessel tree model, the nasal septum mucosa flap model and the preoperative physiological data of the current patient and the historical patients, and obtaining the structural characteristic parameters based on the geometric characteristics of the head and neck blood vessel tree model and the nasal septum mucosa flap model; obtaining the intraoperative mucosa flap operation parameters and the blood vessel perfusion characteristic parameters of each historical patient; determining the reference weight of each historical patient according to the differences between the head and neck blood vessel tree models and the differences between the structural characteristic parameters; training the perfusion prediction model based on the mucosa flap operation parameters, the structural characteristic parameters, the preoperative physiological data and the blood vessel perfusion characteristic parameters of all the historical patients; inputting the artificial planning mucosa flap operation parameters, the structural characteristic parameters and the preoperative physiological data of the current patient into the trained perfusion prediction model, and obtaining the predicted blood vessel perfusion characteristic parameters of the current patient in combination with the reference weight of the historical patients; determining all candidate blood vessel pedicle paths in the head and neck blood vessel tree model of the current patient, and determining the planning blood vessel pedicle path from all the candidate blood vessel pedicle paths according to the predicted blood vessel perfusion characteristic parameters and the change characteristics of each candidate blood vessel pedicle path.

2. The method of preoperative intelligent planning of nasal septum mucosal flap pedicle pathway according to claim 1, characterized in that, The method for obtaining the blood vessel perfusion characteristic parameters comprises: obtaining the ICG perfusion image sequence of the blood vessels of each historical patient during the operation; the ICG perfusion image sequence at least includes the ICG perfusion images of the nasal septum mucosa flap at three operation nodes of flap body separation, pulling and turning over; determining the flap body region in each frame of the ICG perfusion images; in the ICG perfusion image sequence, taking the flap body region with the maximum fluorescence intensity as the target region, taking the frame number corresponding to the target region as the peak time, determining the perfusion speed based on the fluorescence intensity of the target region and the peak time, and determining the perfusion uniformity parameter based on the discrete characteristics of the fluorescence intensity of the target region; taking the vector composed of the peak time, the perfusion speed and the perfusion uniformity parameter as the perfusion characteristic parameter.

3. The method of preoperative intelligent planning of nasal septum mucosal flap pedicle pathway according to claim 1, characterized in that, The method for obtaining the structural characteristic parameters based on the geometric characteristics of the head and neck blood vessel tree model and the nasal septum mucosa flap model comprises: obtaining the blood vessel characteristic parameters in the head and neck blood vessel tree model and the flap body characteristic parameters in the nasal septum mucosa flap model; the blood vessel characteristic parameters at least include the main stem diameter, the distal end minimum diameter, the average bifurcation angle and the total length of the blood vessel tree, and the flap body characteristic parameters at least include the flap body surface area and the flap body thickness; taking the vector composed of the blood vessel characteristic parameters and the flap body characteristic parameters as the structural characteristic parameters.

4. The method of preoperative intelligent planning of nasal septum mucosal flap pedicle pathway according to claim 1, characterized in that, The method for determining the reference weight of each historical patient according to the differences between the head and neck blood vessel tree models and the differences between the structural characteristic parameters comprises: for each head and neck blood vessel tree model, obtaining the blood vessel structure complexity coefficient according to the diameter change of the blood vessel segments corresponding to the adjacent bifurcation points and the bifurcation characteristics of the blood vessels; The tree edit distance between the corresponding head and neck vascular tree models between the current patient and each historical patient is taken as a first difference parameter, the difference between the corresponding structure feature parameters is taken as a second difference parameter, and the difference between the corresponding vascular structure complexity coefficients is taken as a third difference parameter; and the first difference parameter, the second difference parameter and the third difference parameter are fused to obtain a reference weight of each historical patient relative to the current patient.

5. The method of preoperative intelligent planning of nasal septum mucosal flap pedicle pathway according to claim 4, characterized in that, The vascular structure complexity coefficient is obtained according to the diameter change of the corresponding blood vessel segment of the adjacent bifurcation points and the bifurcation characteristics of the blood vessel for each head and neck vascular tree model, and includes: Based on the blood flow direction, the corresponding blood vessel segment of the adjacent bifurcation points in each head and neck vascular tree model is obtained, the bifurcation deformation parameter of each blood vessel segment is obtained according to the diameter change and bending characteristics of each blood vessel segment and the bifurcation angle at the bifurcation point corresponding to the starting point of the blood vessel segment, and the bifurcation deformation parameter of each blood vessel segment is obtained. In the topological structure of each head and neck vascular tree model, the vascular structure complexity coefficient is determined according to the difference between the direction of each blood vessel segment and the remaining blood vessel segments in the same topological level and the bifurcation deformation parameter, and the difference between the branch number corresponding to different nodes in each topological level is combined.

6. The method for preoperative intelligent planning of nasal septum mucosal flap pedicle pathway according to claim 5, characterized in that, The bifurcation deformation parameter is obtained according to the diameter change and bending characteristics of each blood vessel segment and the bifurcation angle at the bifurcation point corresponding to the starting point of the blood vessel segment, and includes: Each blood vessel segment is divided into a predetermined number of subsegments and sorted by label; In each blood vessel segment, the diameter mutation parameter of the blood vessel segment is obtained according to the difference between the diameters corresponding to each subsegment and the adjacent next subsegment; The cosine value of the minimum bifurcation angle at the bifurcation point corresponding to the starting point of each blood vessel segment is taken as the bifurcation influence parameter; The bifurcation deformation parameter is obtained by combining the curvature of the blood vessel segment, the diameter mutation parameter and the bifurcation influence parameter.

7. The method for preoperative intelligent planning of nasal septum mucosal flap pedicle pathway according to claim 5, characterized in that, The vascular structure complexity coefficient is determined according to the difference between the direction of each blood vessel segment and the remaining blood vessel segments in the same topological level and the bifurcation deformation parameter, and the difference between the branch number corresponding to different nodes in each topological level is combined. The topological structure of the head and neck vascular tree model is constructed with the blood vessel bifurcation point as the node and the blood vessel segment as the connection path, and the direction vector of each connection path corresponding to the blood vessel segment is obtained based on the spatial position of the node; The mean value of the sine of the included angle between the direction vector corresponding to the target blood vessel segment and the remaining blood vessel segments in the same topological level is taken as the shape deviation parameter of the target blood vessel segment; The structure complexity parameter of the target blood vessel segment is obtained by combining the bifurcation deformation parameter and the shape deviation parameter; The variation coefficient of the total number of connection paths of all nodes in each topological level is taken as the branch disorder parameter of each topological level; The structure complexity parameters of all blood vessel segments in each topological level are weighted using the branch disorder parameter of each topological level, and the normalized value of the weighted sum result is taken as the vascular structure complexity coefficient of the head and neck vascular tree model.

8. The method for preoperative intelligent planning of nasal septum mucosal flap pedicle pathway according to claim 1, characterized in that, The artificial planning flap operation parameter of the current patient, the structure characteristic parameter and the preoperative physiological data are input into the trained perfusion prediction model, the reference weight of the historical patient is combined, and a predicted blood vessel perfusion characteristic parameter of the current patient is acquired. The flap operation parameter, the structure characteristic parameter and the preoperative physiological data of all historical patients are taken as feature data, the corresponding blood vessel perfusion characteristic parameter is taken as a label of the feature data, sample data is constructed to train a perfusion prediction model; The artificial planning flap operation parameter of the current patient, the structure characteristic parameter and the preoperative physiological data are input into the trained perfusion prediction model, and a blood vessel perfusion characteristic parameter of the current patient is obtained. The reference patient is screened out from all historical patients based on the reference weight, the feature data of the reference patient is input into the trained perfusion prediction model, and a blood vessel perfusion characteristic parameter of the reference patient is obtained. The difference between the blood vessel perfusion characteristic parameter of the current patient and the blood vessel perfusion characteristic parameter of the reference patient is calculated, the difference is weighted by the reference weight of the reference patient, and the mean value is obtained, so that the output deviation of the perfusion prediction model is obtained. The sum of the blood vessel perfusion characteristic parameter of the current patient and the corresponding output deviation is taken as the predicted blood vessel perfusion characteristic parameter of the current patient.

9. The preoperative intelligent planning method of nasal septum mucosal flap pedicle pathway according to claim 7, characterized in that, The planned blood vessel pedicle path is determined from all candidate blood vessel pedicle paths according to the predicted blood vessel perfusion characteristic parameter and the change characteristic of each candidate blood vessel pedicle path, and the planned blood vessel pedicle path is determined from all candidate blood vessel pedicle paths according to the predicted blood vessel perfusion characteristic parameter and the change characteristic of each candidate blood vessel pedicle path. For the current patient, the artificial planning flap operation parameter corresponding to each candidate blood vessel pedicle path is acquired, and the artificial planning flap operation parameter, the structure characteristic parameter and the preoperative physiological data are input into the trained perfusion prediction model to acquire the blood vessel perfusion characteristic parameter output by the model. For each candidate blood vessel pedicle path, the perfusion safety parameter is acquired according to the deviation of the blood vessel perfusion characteristic parameter relative to the predicted blood vessel perfusion characteristic parameter. The path complexity parameter is acquired according to the total length of the candidate blood vessel pedicle path and the structure complexity parameter of each blood vessel segment on the path. The path candidate score is acquired according to the perfusion safety parameter and the path complexity parameter. The planned blood vessel pedicle path is screened out from all candidate blood vessel pedicle paths based on the path candidate score.

10. The method for preoperative intelligent planning of nasal septum mucosal flap pedicle pathway according to claim 1, characterized in that, The flap operation parameter at least includes the pulling length and the turning angle of the nasal septum mucosa flap during each intraoperative operation.

Images