Preoperative human body intelligent positioning system of medical image

Abstract

The invention belongs to the technical field of biomedical engineering, and discloses a preoperative human body intelligent positioning system for medical images, which comprises a data standardization unit for setting a sampling frequency based on a preset respiratory gating signal, collecting multi-modal image data of a patient according to the sampling frequency, and storing the multi-modal image data in a database; processing by using a point location matching method to obtain standardized multi-modal image data; the deformation tracking unit is used for carrying out difference operation and squaring on gray values of different position points in the standardized multi-mode image data to obtain different error squares; combining all error squares to obtain an error square matrix, and generating an error distribution thermodynamic diagram based on the error square matrix; inferring tissue deformation through an error distribution thermodynamic diagram, and updating tissue deformation field parameters by adopting an optimization algorithm; the accuracy and individuation level of preoperative planning are improved, the dynamic assistance and real-time compensation capability can be provided in the operation, and finally high-precision, low-risk and intelligent human body preoperative positioning and navigation are achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of biomedical engineering. More specifically, the present invention relates to a preoperative human body intelligent positioning system for medical images. Background Art

[0002] The patent with the patent publication number CN110711028B discloses a surgical instrument positioning system and its positioning method. The human natural feature information data presented in the preoperative medical images is first input into the positioning system of the present invention, and 3D data reconstruction is performed. Then, the position information of the handheld instrument obtained by the light sensing element is transmitted back to the positioning system. Next, the feature points can be automatically suggested or automatically specified by the positioning system or manually selected by the operator to establish the coordinate relationship between the patient's position and the handheld instrument, so as to solve the influence problem on doctors during the surgical process in the prior art.

[0003] The existing preoperative human body intelligent positioning systems mainly have the following problems:

[0004] Multi-modal image acquisition is not synchronized at the same respiratory phase, resulting in inconsistent CT, MRI, ultrasound and other images in time, which in turn causes registration errors and structural misjudgment. The respiratory gating signal model can only express regular and ideal periodic breathing. In fact, the breathing of patients is often non-sinusoidal in shape, such as asymmetric, slow exhalation and fast inhalation. The model has a large fitting error for the real waveform, and the sampling points cannot be correctly aligned with the physiological phase; the sine model lacks the ability to adjust parameters and cannot flexibly cope with the differences in the respiratory cycle, amplitude and phase of different patients; the fixed CT gray threshold method cannot adapt to the tissue density differences between different individuals and scanning devices, and it is easy to miss or misdetect calcified areas, especially the recognition ability of low-density calcification or areas with blurred boundaries is poor.

[0005] The scanning parameters of different CT devices will affect the absolute value of the CT gray value, causing the CT value of the calcified area to fluctuate up and down. The fixed threshold cannot adapt to multi-device and multi-batch images; during the multi-modal registration process of CT and MRI, due to operations such as respiration, deformation compensation, and interpolation resampling, local blurring or artifacts may appear in the images, interfering with the gray distribution of real calcified tissues, resulting in a decrease in the recognition accuracy of the fixed threshold; during registration or dynamic simulation of anatomical structures, the calcified tissue may undergo spatial deformation, causing changes in density projection. At this time, the fixed CT threshold is likely to miss marginal calcification or misidentify adjacent high-density tissues as calcification. The proportions of soft tissue components (such as fat, muscle, blood vessels, etc.) in different individuals are different, and their CT value means and standard deviations vary greatly. Using a fixed offset coefficient will result in a decrease in the sensitivity of calcification recognition or a decrease in specificity in some populations. The calcified area may show high-density dense calcification, low-density mild calcification, and scattered punctate or blurred boundary calcification. Summary of the Invention

[0006] To overcome the above-mentioned defects of the prior art and to achieve the above object, the present invention provides the following technical solutions: A preoperative human body intelligent positioning system for medical images, comprising:

[0007] A data standardization unit, which sets a sampling frequency based on a preset respiratory gating signal, acquires multi-modal image data of a patient according to the sampling frequency, and processes the data using a point matching method to obtain standardized multi-modal image data;

[0008] A deformation tracking unit, which performs a difference operation and squares the gray values of different position points in the standardized multi-modal image data to obtain different squared errors; combines all the squared errors to obtain a squared error matrix, generates an error distribution heat map based on the squared error matrix; and infers tissue deformation through the error distribution heat map, and updates tissue deformation field parameters using an optimization algorithm;

[0009] An attenuation rule construction unit, which generates a tissue layer mask based on the standardized multi-modal image data and the updated tissue deformation field parameters, assigns a preset basic attenuation coefficient, and constructs a multi-modal attenuation rule table;

[0010] A risk avoidance unit, which performs surgical path planning based on the tissue layer mask and the multi-modal attenuation rule table, respectively identifies the positions of the calcified layer and the large blood vessels and performs surgical avoidance to obtain an optimized surgical path;

[0011] A navigation and positioning unit, which generates an AR navigation interface by combining the optimized surgical path and a preset instrument space coordinate mapping table; based on the AR navigation interface, performs preoperative human body intelligent positioning through dynamic elastic compensation and multi-modal path matching.

[0012] Preferably, the multi-modal image data includes CT images, MRI images, and ultrasound images.

[0013] Preferably, the method for obtaining the standardized multi-modal image data includes:

[0014] Determine the sampling frequency of a unified image acquisition device based on a respiratory gating signal, acquire the real-time respiratory signal of the patient, and select the end of inspiration as the trigger reference phase for unified sampling; construct a respiratory gating signal model based on the real-time respiratory signal of the patient and calibrate the key phase points in the respiratory cycle;

[0015] In two consecutive respiratory cycles, trigger the image acquisition device to acquire CT images, MRI images, and ultrasound images at the same respiratory phase; set the sampling frequency of the image acquisition device to an integer multiple of the patient's respiratory frequency, and align the acquisition time points of the image acquisition device with the respiratory phase;

[0016] Taking the CT image as the registration reference, m marker points in the CT image are extracted to construct a CT reference coordinate system, and the SURF feature matching method is used to identify the corresponding marker point sets in the MRI image and the ultrasound image; the iterative closest point algorithm is used to align different marker points in the MRI image and the ultrasound image with the corresponding marker points in the CT reference coordinate system respectively, so as to obtain the registered multimodal image data;

[0017] Perform resolution unification processing on the multimodal image data, preset a unified target resolution, use bilinear interpolation for the ultrasound image in the multimodal image data, and use trilinear interpolation for the CT image and the MRI image in the multimodal image data for resampling to make it reach the preset unified target resolution, thereby obtaining the multimodal image data after the unified target resolution;

[0018] Calculate the gray values of the multimodal image data after resolution unification, obtain the minimum gray value and the maximum gray value of each pixel in the CT image, the MRI image and the ultrasound image, perform linear transformation normalization processing on the minimum gray value and the maximum gray value of each pixel, and obtain the normalized multimodal image data; redistribute the gray value distribution of each image in the normalized multimodal image data by the histogram equalization method, thereby obtaining the standardized multimodal image data.

[0019] Preferably, the method for obtaining the error distribution heat map includes:

[0020] Extract the CT image and the MRI image from the standardized multimodal image data, extract the corresponding gray values for each voxel position point in the CT image and the MRI image, subtract and square the gray values to obtain different squared errors, and combine the squared errors at all voxel positions in the CT image and the MRI image into a squared error matrix; sum the squared errors at all voxel position points, then divide by the total number of voxel position points to obtain the average squared error, and take the square root of the average squared error to obtain the root mean square error value;

[0021] Preset a root mean square error value threshold. If the root mean square error value is greater than or equal to the error value threshold, it is determined that the gray value difference between the CT image and the MRI image is abnormal and directly marked; if the root mean square error value is less than the preset root mean square error value threshold, it is determined that the gray value difference between the CT image and the MRI image is normal, and the error distribution heat map starts to be generated; slice the squared error matrix layer by layer, each layer represents the error heat map in one direction, normalize the gray value error range of each layer, map the gray value error to the color space, and different gray value error sizes will correspond to different colors, and combine the error heat maps of all slices one by one to form the error distribution heat map.

[0022] Preferably, the method for updating the elastic deformation field parameters includes:

[0023] A gray value error threshold and a voxel position point quantity threshold are preset. If the gray value error of the voxel position points in any area of ​​the error distribution heat map is greater than or equal to the preset gray value error threshold and greater than the preset voxel position point quantity threshold, the area is determined to be a high-risk area. If the gray value error of the voxel position points in any area of ​​the error distribution heat map is less than the preset gray value error threshold and greater than the preset voxel position point quantity threshold, the area is determined to be a low-risk area. In the error distribution heat map, the high-risk area is represented by red, and the low-risk area is represented by green.

[0024] For high-risk areas, the corresponding image areas in the standardized multimodal image data are extracted, and the feature points are matched between continuous time frames using the SIFT feature matching algorithm to obtain matching results. Based on the matching results, the tissue displacement in the high-risk area is calculated.

[0025] The respiratory phases are divided according to the respiratory gating signal model. In one respiratory phase, the end of exhalation is taken as the reference phase, and the tissue displacement of the inhalation phase relative to the reference phase is calculated to form a dynamic displacement table for each area. The dynamic offsets of all areas are summarized with the tissue displacements in the high-risk areas to generate a tissue deformation displacement table. According to the tissue deformation displacement table, the tissue deformation field parameters are updated through the gradient descent optimization algorithm to obtain the updated tissue deformation field parameters.

[0026] Preferably, the method for constructing a multi-modal attenuation rule table comprises:

[0027] For the tissue deformation field of each frame of CT images and MRI images, the average displacement of all voxels is calculated as the motion intensity index of the frame to obtain the motion intensity of each frame; a motion stability threshold is preset, the motion intensity of all frames is compared with the preset motion stability threshold, and all frames with motion intensity less than the preset motion stability threshold are screened out as reference frame sets; all reference frame sets are uniformly mapped to the coordinate system of the frame through non-rigid transformation to form a deformation compensation reference coordinate system; based on the updated tissue deformation field parameters, spatial coordinate mapping is performed on the CT images and MRI images in the standardized multimodal image data set to be unified to the deformation compensation reference coordinate system;

[0028] The image segmentation algorithm is used to segment the CT images and MRI images after deformation compensation, identify the target organs, and generate a three-dimensional mask corresponding to each target organ; based on the updated tissue deformation field parameters, the tissue structure within the three-dimensional mask region of different target organs is divided, and according to the image feature parameters and the preset tissue feature template, different tissue types of the target organs are identified, and the corresponding tissue hierarchical mask is constructed; the image feature parameters include CT value, MRI signal intensity, and T2 relaxation time difference; the position and shape of the tissue hierarchical mask are adjusted through the B-spline elastic deformation field to generate a fine tissue hierarchical mask;

[0029] Based on the different tissue types of the identified target organs, the corresponding preset basic attenuation coefficients are assigned, and the tissue classification threshold is set in combination with the image feature parameters. The basic attenuation coefficients corresponding to all tissue types of the target organs are summarized to form a basic attenuation coefficient table, and a list of basic attenuation coefficients corresponding to each tissue type is obtained; the basic attenuation coefficients of all tissue types are combined with the fine tissue hierarchical mask to form a multi-modal attenuation rule table.

[0030] Preferably, the method for obtaining the optimized surgical path includes:

[0031] Extract the CT gray value CT(x,y,z) and the corresponding tissue label La(x,y,z) at each voxel position in the tissue hierarchical mask; preset the calcified CT threshold as T ca , define the mask function of the calcified layer, and determine whether the voxel point at the voxel position (x,y,z) belongs to the calcified layer region through the mask function of the calcified layer;

[0032] Dynamically adjust the calcification threshold through the threshold adjustment function, and dynamically adjust the offset coefficient in the threshold adjustment function through the offset coefficient adjustment function; preset the large blood vessel anatomical region template, and based on the preset large blood vessel anatomical region template and the fine tissue hierarchical mask, perform position and shape matching on the region to be identified to identify the large blood vessel region; and mark the calcified layer region as a no-go area and the large blood vessel region as a core area;

[0033] Set the surgical target point and the surgical starting point, generate a preliminary surgical path, and stipulate that the preliminary surgical path must avoid the no-go area and the core area. Calculate the minimum cost path avoiding the no-go area and the core area through the Dijkstra algorithm, and then obtain the optimized surgical path.

[0034] Preferably, the method for generating the AR navigation interface includes:

[0035] A preset instrument space coordinate mapping table, based on standardized multimodal imaging data, performs tissue segmentation and standardization on the target organ, differentiates different tissue types of the target organ with different colors or transparencies, uses the Marching Cubes algorithm to generate a three-dimensional anatomical structure model of the patient, superimposes the optimized surgical path in the form of lines or arrows on the patient's three-dimensional anatomical structure model, and according to the preset instrument space coordinate mapping table, updates and displays the position and orientation of the surgical instrument in the three-dimensional model in real time, and generates an AR navigation interface.

[0036] Preferably, the method for performing preoperative human body intelligent positioning includes

[0037] According to the updated tissue deformation field parameters, extract the dynamic deformation information of the tissue at different time points, adjust the patient's three-dimensional anatomical model and the optimized surgical path in real time, and combine the respiratory gating signal model with the updated tissue deformation field parameters; based on CT images, MRI images and ultrasound images, generate a surgical path matching table, and compare whether the current optimized surgical path deviates from the path in the surgical path matching table through the surgical path matching table. If there is a deviation, then readjust the current optimized surgical path according to the surgical path matching table to obtain a high-precision optimized surgical path;

[0038] Based on the three-dimensional anatomical model, mark the key anatomical landmark points, obtain the initial position of the surgical instrument, map the initial position of the surgical instrument onto the patient's three-dimensional anatomical model, align it with the key anatomical landmark points, and through the results of dynamic elastic compensation and multimodal path matching, dynamically adjust the spatial relationship between the surgical instrument and the target position to perform preoperative human body intelligent positioning.

[0039] Preferably, the method for dynamic elastic compensation includes:

[0040] Obtain the intraoperative real-time imaging data of the patient, and extract the dynamic deformation characteristics of the tissue in combination with the intraoperative real-time imaging data; construct an elastic deformation field based on a preset biomechanical model, and the preset biomechanical model includes the elastic modulus and deformation constraint conditions of different tissues;

[0041] Generate a dynamic displacement vector field according to the elastic deformation field, and map it into the patient's three-dimensional anatomical structure model in real time to correct the deformation error of the three-dimensional anatomical structure model caused by physiological movement; based on the corrected three-dimensional anatomical structure model and the high-precision optimized surgical path, dynamically adjust the position and running trajectory of the surgical instrument to align the surgical instrument with the target position.

[0042] The technical effects and advantages of a preoperative human body intelligent positioning system for medical images according to the present invention:

[0043] By independently modeling the patient's breathing process in two stages: inhalation and exhalation, the present invention can more accurately simulate the asymmetric breathing movement of the patient; this segmented modeling not only improves the precise positioning of key phase points such as the end of inhalation and the end of exhalation, but also avoids the phase shift error caused by the traditional sine wave model during the acquisition process; by setting the sampling frequency of the imaging acquisition device to an integer multiple of the patient's breathing frequency and performing multi-modal imaging acquisition at the same breathing phase, it is ensured that CT images, MRI images, and ultrasound images are acquired at the same physiological moment, thereby eliminating the time misalignment between different imaging modalities;

[0044] Adopting the setting of dynamic calcification thresholds based on the statistical characteristics of soft tissue regions can adapt to the tissue density differences of different individuals, thus effectively avoiding misjudgment, and flexibly adjusting according to the specific physiological conditions of the patient, the scanning device, and different imaging conditions, improving the accuracy of calcification region recognition, especially greatly enhancing the recognition ability of low-density calcification or calcification with blurred boundaries; by introducing the neighborhood standard deviation and combining the statistical comparison between the local region and the adjacent region, it can improve the stability and accuracy of calcification recognition when the image noise is large or the tissue boundary is blurred; it has a strong recognition ability for calcification regions with blurred boundaries, effectively avoiding the false detection problems easily occurring in traditional methods. By dynamically adjusting the offset coefficient, it can adapt to the soft tissue density fluctuations of different patients and optimize the judgment criteria for calcification regions; by accurately marking the calcification layer as the no-go area and the large blood vessel area as the core area, and generating the minimum cost path that avoids the no-go area and the core area, it is ensured that the surgical path avoids dangerous areas, improving the safety and feasibility of the surgery; the optimized surgical path not only avoids possible calcification areas, but also can effectively avoid large blood vessel areas, thereby reducing the surgical risk. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a schematic structural diagram of a pre-operative human body intelligent positioning system for medical images of the present invention;

[0046] Figure 2 It is a schematic flow diagram of a pre-operative human body intelligent positioning method for medical images of the present invention;

[0047] Figure 3 It is a schematic flow diagram of a method for obtaining a heat map of error distribution provided by the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0048] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0049] Example 1

[0050] Please refer to Figure 1 and Figure 3 as shown in the figure. This Example 1 further illustrates a preoperative human body intelligent positioning system for medical images proposed by the present invention, including:

[0051] In the process of surgical planning and intraoperative navigation guided by medical images, relying on multimodal images (such as CT, MRI, ultrasound, etc.) to construct a three-dimensional anatomical model of the patient and realizing lesion area recognition and high-risk tissue avoidance have become the key supports for the development of intelligent and precise surgery. However, during the current multimodal image acquisition process, due to the lack of synchronization with the unified respiratory phase, the inconsistency of different modal images on the time axis often occurs, which in turn leads to the mismatch of organ morphology and position, affects the registration accuracy, and even causes misjudgment of structure recognition, seriously restricting the accuracy of preoperative evaluation and path planning.

[0052] Traditional respiratory gating techniques mostly use idealized sine functions or regular periodic models to approximate the patient's breathing process for guiding image sampling and physiological phase alignment. However, in actual clinical practice, the patient's breathing often shows non-ideal waveform characteristics, such as fast inhalation, slow exhalation, asymmetry, and even unstable rhythm. The standard sine model is difficult to accurately fit the real waveform, resulting in misalignment of sampling points and large deviations in tissue motion estimation. In addition, the parameters of the conventional sine model are fixed, lacking the adaptability to the breathing cycle, breathing amplitude, and phase changes of different patients, affecting the temporal consistency and spatial matching accuracy of image registration.

[0053] In terms of the identification of calcified tissues, at present, many solutions still use fixed CT gray value thresholds for screening. However, there are differences in the proportions of soft tissue components (such as fat, muscle, blood vessels, etc.) among different individuals, and there are also obvious individual differences in the mean value and standard deviation of their CT gray value distributions. The unified threshold strategy is prone to missed detection or false detection of low-density calcifications and calcifications with blurred edges, especially in complex soft tissue backgrounds, with insufficient specificity and sensitivity. At the same time, the scanning parameters of different CT devices (such as tube voltage, current, reconstruction algorithm, etc.) will affect the absolute numerical range of CT values, and there are CT gray value fluctuations in images collected across devices and time points. The fixed threshold strategy is difficult to stably adapt to multi-batch data, resulting in a significant decline in the recognition accuracy.

[0054] In addition, during the multi-modal image registration process of CT and MRI, local blurring or artifacts may be introduced due to interpolation resampling, deformation compensation, and image alignment processing. These non-genuine image information interfere with the judgment of the gray-scale distribution in the calcified area, making the traditional recognition method based on fixed CT values lack robustness. Calcified tissues may also undergo a certain degree of spatial deformation during dynamic registration and anatomical reconstruction, which in turn affects their density distribution performance in the projection image, easily leading to the failure to identify the edge calcified area or misjudging adjacent high-density tissues as calcified areas.

[0055] To effectively solve the above problems, the present invention proposes a preoperative human body intelligent positioning system for medical images, including:

[0056] A data standardization unit sets the sampling frequency based on a preset respiratory gating signal, acquires multi-modal image data of a patient according to the sampling frequency, and processes it using a point matching method to obtain standardized multi-modal image data;

[0057] A deformation tracking unit performs difference operations and squares on the gray-scale values of different position points in the standardized multi-modal image data to obtain different squared errors; combines all the squared errors to obtain a squared error matrix, generates an error distribution heat map based on the squared error matrix; and infers tissue deformation through the error distribution heat map, and updates the tissue deformation field parameters using an optimization algorithm;

[0058] An attenuation rule construction unit generates a tissue layer mask based on the standardized multi-modal image data and the updated tissue deformation field parameters, assigns a preset basic attenuation coefficient, and constructs a multi-modal attenuation rule table;

[0059] A risk avoidance unit performs surgical path planning based on the tissue layer mask and the multi-modal attenuation rule table, respectively identifies the positions of the calcified layer and the large blood vessels and performs surgical avoidance to obtain an optimized surgical path;

[0060] A navigation positioning unit generates an AR navigation interface by combining the optimized surgical path and a preset instrument space coordinate mapping table; based on the AR navigation interface, performs preoperative human body intelligent positioning through dynamic elastic compensation and multi-modal path matching.

[0061] The multi-modal image data includes CT images, MRI images, and ultrasound images.

[0062] The imaging principle of CT images is that CT scanning uses X-rays to pass through different layers of the human body, combines computer processing, and generates cross-sectional images of organs and tissues, which can provide high-resolution bone structure images and can also be used to view soft tissues and other tissues with large density differences. The resolution of CT images is usually relatively high, especially for bone tissues.

[0063] The imaging principle of MRI is that MRI uses a strong magnetic field and radiofrequency waves to generate detailed images of different tissues in the body. By imaging the differences in the responses of water molecules in different tissues. MRI images are radiation-free and can generate very fine images of soft tissue structures. Especially in the imaging of soft tissues, MRI is superior to CT and can well distinguish different types of soft tissues such as fat, muscle, and brain tissue.

[0064] The imaging principle of ultrasound imaging is that when ultrasonic waves pass through tissues in the body, they will be reflected, and the reflected waves are processed to generate images. It is an imaging technology based on sound waves and is usually used for real-time image acquisition. Ultrasound images are radiation-free and can perform real-time imaging, which is suitable for dynamic observation. Ultrasound images are usually used to visualize the morphology of liquids, tissues, and organs in the body. It is sensitive to the reflected waves of different tissues and can display the movement of muscles, fat, organs, blood flow, etc.

[0065] The methods for obtaining standardized multi-modal image data include:

[0066] Determine the sampling frequency of the unified image acquisition device based on the respiratory gating signal, collect the patient's real-time respiratory signal, and select the end of inspiration as the trigger reference phase for unified sampling; construct a respiratory gating signal model based on the patient's real-time respiratory signal and calibrate the key phase points in the respiratory cycle;

[0067] The respiratory gating signal model is R(t) = A(t)·sin(2πf r t + φ); where, R(t) represents the patient's real-time respiratory signal at time point t; A(t) represents the respiratory amplitude, corresponding to the maximum change in the movement of the chest or abdomen during the patient's breathing process; f r represents the respiratory frequency, representing the number of respiratory cycles of the patient; t represents the index of the time point; φ represents the initial phase, which is used to align the acquisition starting point. For example, if the start of inspiration is used as the reference point for t = 0, then φ = 0 can be set; 2πf r t + φ represents the phase angle during the breathing process;

[0068] However, the respiratory gating signal model can only express regular and ideal periodic breathing. In fact, the patient's breathing is often non-sinusoidal in shape, such as asymmetric, slow exhalation and fast inhalation. The model has a large fitting error for the real waveform, and the sampling points cannot be correctly aligned with the physiological phase; the sine model lacks the ability to adjust parameters and cannot flexibly handle the differences in the respiratory cycle, amplitude, and phase of different patients;

[0069] To solve the above asymmetry problem in the patient's respiratory cycle, the respiratory amplitude A(t) is defined as a piecewise function of the inhalation phase and the exhalation phase; where, A in (t) represents the regulation function in the inhalation phase; t in represents the time point when inhalation starts; tex Indicates the respiratory phase switching point, representing the boundary time point between inhalation and exhalation; A ex r(t) represents the exhalation phase adjustment function; t ed Represents the time point when exhalation ends;

[0070] Inhalation phase adjustment function Among them, α represents the inhalation adjustment parameter, which determines the growth rate of the inhalation phase;

[0071] Exhalation phase adjustment function A max Represents the maximum displacement amplitude within a complete respiratory cycle;

[0072] It should be noted that the natural respiratory cycle of the human body itself is asymmetric and usually shows: inhalation is faster (active muscle contraction), and exhalation is slower (passive rebound); that is, a standard respiratory cycle is not a symmetric sine wave but more closely resembles a "skewed sawtooth wave" or "asymmetric bell-shaped wave". The inhalation phase is usually an active process, driven by the contraction of the diaphragm and the expansion of the intercostal muscles, with a faster intake speed and a rapid increase in lung volume; the exhalation phase is mostly a passive process, completed by the elastic recoil of lung tissue and the chest wall, with a slower exhalation speed and a slow decrease in volume.

[0073] Fitting modeling is carried out for this respiratory asymmetry. A complex physiological signal is segmented into two independently adjustable functions, which can more realistically simulate this asymmetric respiratory movement law, significantly superior to the traditional model that uses a single sine wave to describe the entire respiratory cycle. The inhalation phase adjustment function can simulate the process of rapid increase in respiratory amplitude in a short time, with adjustable shape and smooth boundary, suitable for fitting the rapid dynamic changes during inhalation; the exhalation phase adjustment function: simulates a slow and smooth decay process, matching the natural rhythm of passive recoil, with good continuity and differentiability, ensuring calculation stability. By reasonably fitting the actual respiratory curve, key physiological phases (such as the end of inhalation and the end of exhalation) can be predicted more accurately, used as the trigger point for image acquisition, improving the spatial and temporal consistency between images. Avoiding the phase shift error caused by using the ideal sine wave model and reducing image artifacts and registration errors caused by sampling misalignment.

[0074] Advantages over the prior art: The breathing process is modeled in two segments, an inhalation function and an exhalation function; the inhalation rising rate and the exhalation descending curve are adjusted separately to truly restore the "fast inhalation and slow exhalation" rhythm; the positioning accuracy of the breathing phase points (end of inhalation, end of exhalation) is improved, especially suitable for gated trigger sampling. Accurately fitting the non-linear phase curve of each cycle enables the image acquisition device to sample under the true physiological rhythm; solving problems such as image blurring and tissue misalignment caused by incorrect phase triggering. Synchronizing the multi-modal image acquisition points at the true end of inhalation (or end of exhalation) can avoid spatial structure misalignment; providing a time-based guarantee for subsequent steps such as SURF feature matching and ICP iterative registration.

[0075] In two consecutive breathing cycles, trigger the imaging acquisition device to acquire CT images, MRI images, and ultrasound images at the same breathing phase; set the sampling frequency of the imaging acquisition device to an integer multiple f of the patient's breathing frequency s = N·f r ; where, f s represents the sampling frequency of the imaging acquisition device; N represents the number of times the acquisition device samples in one breathing cycle; align the acquisition time points of the imaging acquisition device with the breathing phase;

[0076] Taking the CT image as the registration reference, extract m marker points in the CT image to construct a CT reference coordinate system, and use the SURF feature matching method to identify the corresponding marker point sets in the MRI image and the ultrasound image; through the iterative closest point algorithm, align the different marker points in the MRI image and the ultrasound image with the corresponding marker points in the CT reference coordinate system respectively to obtain the registered multi-modal image data;

[0077] Perform resolution unification processing on the multi-modal image data, preset a unified target resolution, perform bilinear interpolation on the ultrasound image in the multi-modal image data, and perform trilinear interpolation on the CT image and the MRI image in the multi-modal image data for resampling to make it reach the preset unified target resolution, thereby obtaining the multi-modal image data after the unified target resolution;

[0078] Calculate the gray values of the multi-modal image data after resolution unification, obtain the minimum gray value and the maximum gray value of each pixel in the CT image, the MRI image, and the ultrasound image, perform linear transformation normalization processing on the minimum gray value and the maximum gray value of each pixel to obtain the normalized multi-modal image data; redistribute the gray value distribution of each image in the normalized multi-modal image data through the histogram equalization method to obtain the standardized multi-modal image data.

[0079] The method for obtaining the error distribution heat map includes:

[0080] Extract the CT images and MRI images from the standardized multimodal image data. For each voxel position point in the CT images and MRI images, extract the corresponding gray value, subtract the gray values and square the result to obtain different squared errors. Combine the squared errors at all voxel positions in the CT images and MRI images into a squared error matrix; sum up the squared errors at all voxel position points, then divide by the total number of voxel position points to obtain the mean squared error, and take the square root of the mean squared error to obtain the root mean square error value.

[0081] It should be noted that when performing error analysis and gray value contrast evaluation, only the gray values in the CT images and MRI images are selected for comparison, excluding ultrasound images; the main reason is that the gray value of the CT image represents the degree of X-ray attenuation (i.e., tissue density), which is a quantitative image; the gray value of the MRI image represents the response of the tissue to the magnetic field and radiofrequency pulse. Although it is qualitative, it still has relative consistency after standardization; while the gray value of the ultrasound image reflects the echo intensity, which is affected by multiple factors such as tissue interface, depth, direction, scattering, and noise, and the gray values of the same structure under different probes and angles vary greatly. The CT images and MRI images are 3D volume images, and voxel alignment can be achieved. The ultrasound image is often a 2D slice or collected by a freehand probe. Even if it is reconstructed into 3D, its spatial information is not as accurate as that of the CT images and MRI images; factors such as spatial resolution, acquisition angle, and soft tissue deformation make it difficult to achieve one-to-one correspondence of voxels even after the ultrasound image is registered.

[0082] Preset a root mean square error value threshold. If the root mean square error value is greater than or equal to the error value threshold, it is determined that the gray value difference between the CT image and the MRI image is abnormal and directly marked; if the root mean square error value is less than the preset root mean square error value threshold, it is determined that the gray value difference between the CT image and the MRI image is normal, and the error distribution heat map starts to be generated; slice the squared error matrix layer by layer. Each layer represents the error heat map in a direction (such as the transverse plane or the coronal plane). Normalize the gray value error range of each layer so that the gray value error is mapped into the color space. Different gray value error magnitudes will correspond to different colors. Combine the error heat maps of all slices one by one to form the error distribution heat map.

[0083] The method for updating the elastic deformation field parameters includes:

[0084] A gray value error threshold and a voxel position point quantity threshold are preset. If the gray value error of the voxel position points in any area of ​​the error distribution heat map is greater than or equal to the preset gray value error threshold and greater than the preset voxel position point quantity threshold, the area is determined to be a high-risk area. If the gray value error of the voxel position points in any area of ​​the error distribution heat map is less than the preset gray value error threshold and greater than the preset voxel position point quantity threshold, the area is determined to be a low-risk area. In the error distribution heat map, the high-risk area is represented by red, and the low-risk area is represented by green.

[0085] For high-risk areas, the corresponding image areas in the standardized multimodal image data are extracted, and the feature points are matched between continuous time frames using the SIFT feature matching algorithm to obtain matching results. Based on the matching results, the tissue displacement in the high-risk area is calculated.

[0086] The respiratory phases are divided according to the respiratory gating signal model. In one respiratory phase, the end of exhalation is taken as the reference phase, and the tissue displacement of the inhalation phase relative to the reference phase is calculated to form a dynamic displacement table for each area. The dynamic offsets of all areas are summarized with the tissue displacements in the high-risk areas to generate a tissue deformation displacement table. According to the tissue deformation displacement table, the tissue deformation field parameters are updated through the gradient descent optimization algorithm to obtain the updated tissue deformation field parameters.

[0087] The method of constructing a multi-modal attenuation rule table includes:

[0088] For the tissue deformation field of each frame of CT images and MRI images, the average displacement of all voxels is calculated as the motion intensity index of the frame to obtain the motion intensity of each frame; a motion stability threshold is preset, the motion intensity of all frames is compared with the preset motion stability threshold, and all frames with motion intensity less than the preset motion stability threshold are screened out as reference frame sets; all reference frame sets are uniformly mapped to the coordinate system of the frame through non-rigid transformation to form a deformation compensation reference coordinate system; based on the updated tissue deformation field parameters, spatial coordinate mapping is performed on the CT images and MRI images in the standardized multimodal image data set to be unified to the deformation compensation reference coordinate system;

[0089] The image segmentation algorithm is used to segment the CT images and MRI images after deformation compensation, identify the target organs, and generate a three-dimensional mask corresponding to each target organ; based on the updated tissue deformation field parameters, the tissue structures within the three-dimensional mask regions of different target organs are divided, and according to the image feature parameters and the preset tissue feature templates, different tissue types of the target organs (such as different tissue types like fat, muscle, blood vessels, etc.) are identified, and the corresponding tissue layered masks are constructed; the image feature parameters include CT value, MRI signal intensity, and T2 relaxation time difference; the position and shape of the tissue layered mask are adjusted through the B-spline elastic deformation field to generate a fine tissue layered mask;

[0090] Based on the different tissue types of the identified target organs, the corresponding preset basic attenuation coefficients are assigned, and the tissue classification thresholds are set in combination with the image feature parameters. The basic attenuation coefficients corresponding to all tissue types of the target organs are summarized to form a basic attenuation coefficient table, and a list of basic attenuation coefficients corresponding to each tissue type is obtained; the basic attenuation coefficients of all tissue types are combined with the fine tissue layered mask to form a multi-modal attenuation rule table.

[0091] The method for obtaining an optimized surgical path includes:

[0092] Extract the CT gray value CT(x, y, z) and the corresponding tissue label La(x, y, z) at each voxel position in the tissue layered mask; the preset calcification CT threshold is T ca , define the mask function of the calcified layer, and determine whether the voxel point at the voxel position (x, y, z) belongs to the calcified layer region through the mask function of the calcified layer; define the mask function of the calcified layer as where Mca(x, y, z) = 1 indicates that the voxel point at the voxel position (x, y, z) belongs to the calcified layer region; Mca(x, y, z) = 0 indicates that the voxel point at the voxel position (x, y, z) does not belong to the calcified layer region;

[0093] In clinical image analysis and surgical path planning based on CT images, the accurate identification of calcified regions is a crucial step to ensure safe operations. Traditional methods usually use a fixed CT gray-scale threshold (e.g., >130 HU) to identify calcified tissues. However, there are significant differences in the physical constitutions, organ densities, and calcification degrees of different patients. The CT value distribution of the same tissue may shift among different individuals, resulting in missed or misdetected cases with a fixed threshold. The scanning parameters of different CT devices (such as kVp, current, and reconstruction algorithms) can affect the absolute values of CT gray-scale values, causing fluctuations in the CT values of calcified regions. A fixed threshold cannot adapt to images from multiple devices and batches. During the multi-modal registration of CT and MRI, due to operations such as respiration, deformation compensation, and interpolation resampling, local blurring or artifacts may appear in the images, interfering with the gray-scale distribution of real calcified tissues and reducing the recognition accuracy of the fixed threshold. In registration or dynamic simulation of anatomical structures, calcified tissues may undergo spatial deformation, causing changes in density projection. At this time, a fixed CT threshold is likely to miss marginal calcifications or misidentify high-density adjacent tissues as calcifications.

[0094] Dynamically adjust the calcification threshold through a threshold adjustment function, and the threshold adjustment function is T ca = μ so + k ca ·σ so ; where μ so represents the average CT gray-scale value within the soft tissue region; σ so represents the standard deviation of CT gray-scale values within the soft tissue region; k ca represents the offset coefficient, indicating the offset amplitude of the calcification recognition threshold relative to the average CT gray-scale value within the soft tissue region;

[0095] It should be noted that the soft tissue region is the reference region. μ so provides a background gray-scale benchmark to help determine whether the voxel at a certain voxel position (x, y, z) belongs to the calcified layer region. The density of calcified tissues is much higher than that of soft tissues, so its CT value is significantly higher than μ so . Dynamically adjusting the calcification threshold through the threshold adjustment function is equivalent to constructing a dynamic recognition boundary. Voxels with CT values higher than the calcification threshold are determined to be in the calcified layer region;

[0096] Threshold adjustment function: An adaptive threshold setting method based on statistical distribution, similar to outlier detection based on Gaussian distribution; there may be significant individual differences in the soft tissue density (CT value) of different patients, especially more obvious in the elderly, disease states, and different scanning devices. If a fixed CT value (such as 130 HU) is used to judge calcification, it is easy to cause misjudgment (missed detection or false detection). By extracting the mean and standard deviation of the current soft tissue area, it can dynamically adapt to the tissue distribution characteristics under different individuals or scanning conditions, improving the generalization ability and accuracy; the beneficial effects compared to the prior art: introducing the statistical characteristics (mean and standard deviation) of local soft tissues, the threshold can be dynamically set according to the actual image gray level, adapting to the tissue density distribution of different individuals, effectively avoiding missed detection or false detection caused by a fixed threshold; in the case of mild calcification degree or blurred boundary, the potential calcification area can still be determined through the local abnormality degree, enhancing the recognition ability; realizing the autonomous detection of the calcification area and the delineation of the surgical prohibited area, greatly reducing the dependence on manual discrimination and improving the intelligence level of the system; avoiding penetration difficulties, bleeding risks or navigation errors caused by ignoring calcification, thereby enhancing the accessibility and safety of the overall surgical path;

[0097] The proportions of soft tissue components (such as fat, muscle, blood vessels, etc.) in different individuals are different, and their mean CT value μ so and standard deviation σ so have large differences; if a fixed offset coefficient is used, it will lead to a decrease in the sensitivity of calcification recognition (inability to recognize low-density calcification) in some populations, or a decrease in specificity (misjudging high-density soft tissue as calcification); the calcification area may show high-density dense calcification, low-density mild calcification, and scattered punctate or blurred boundary calcification; under different morphologies, their CT value distributions also have differences, and only by flexibly adjusting the offset coefficient can the recognition be accurately set;

[0098] Dynamically adjust the offset coefficient in the threshold adjustment function through the offset coefficient adjustment function. The offset coefficient adjustment function where, μ re represents the average CT gray value of all voxels in the current area to be recognized; μ ne represents the average CT gray value of all voxels in the adjacent area of the current area to be recognized; σ ne represents the standard deviation of the CT gray values of all voxels in the adjacent area of the current area to be recognized; λ represents the sensitivity adjustment factor, controlling the adjustment intensity of the offset coefficient. According to the expert experience method, λ ∈ (0, 1]; the current area to be recognized represents the tissue area to be recognized where calcification may exist;

[0099] It should be noted that the offset coefficient adjustment function is similar to the idea of Z-score normalization. If there is a significant difference between the mean value of the candidate region and the surrounding tissues (for example, calcification is much brighter than soft tissue), this ratio will be larger, indicating obvious local abnormal intensity; introducing the neighborhood standard deviation as a normalization term helps to avoid errors when the image has high noise or the tissue boundary is blurred;

[0100] The beneficial effects compared with the prior art are as follows: By adopting the dynamic CT threshold adjustment method based on regional contrast enhancement, introducing the comparison of the local region average value and the statistical characteristics of adjacent tissues, it can achieve a more sensitive judgment of the calcification region and effectively detect mild calcification or calcification structures with blurred boundaries; By dynamically adjusting the offset coefficient and binding it to the statistical information of the current region and its adjacent regions, the judgment criteria can be adaptively adjusted according to the soft tissue density fluctuations of different individuals;

[0101] Preset the anatomical region template of large blood vessels. Based on the preset anatomical region template of large blood vessels and the fine tissue stratification mask, perform position and shape matching on the region to be recognized to identify the large blood vessel region; And mark the calcification layer region as a no-go area and mark the large blood vessel region as the core area;

[0102] Set the surgical target point and the surgical starting point, generate a preliminary surgical path, and stipulate that the preliminary surgical path must avoid the no-go area and the core area. Calculate the minimum cost path that avoids the no-go area and the core area through the Dijkstra algorithm, and then obtain the optimized surgical path.

[0103] The method for generating an AR navigation interface includes:

[0104] Preset the instrument space coordinate mapping table. Based on the standardized multimodal image data, perform tissue segmentation and standardization on the target organ, distinguish different tissue types of the target organ, such as bones, soft tissues, and blood vessels, with different colors or transparencies, use the Marching Cubes algorithm to generate a three-dimensional anatomical structure model of the patient, superimpose the optimized surgical path on the three-dimensional anatomical structure model of the patient in the form of lines or arrows, and according to the preset instrument space coordinate mapping table, update and display the position and direction of the surgical instrument in the three-dimensional model in real time, and generate an AR navigation interface.

[0105] The method for performing preoperative human body intelligent positioning includes

[0106] Extract the dynamic deformation information of the tissue at different time points according to the updated tissue deformation field parameters, and adjust the patient's three-dimensional anatomical model and optimize the surgical path in real time. Combine the respiratory gating signal model with the updated tissue deformation field parameters; generate a surgical path matching table based on CT images, MRI images, and ultrasound images. Compare whether there is a deviation between the current optimized surgical path and the path in the surgical path matching table. If there is a deviation, readjust the current optimized surgical path according to the surgical path matching table to obtain a high-precision optimized surgical path;

[0107] Based on the three-dimensional anatomical model, mark the key anatomical landmark points, obtain the initial position of the surgical instrument, map the initial position of the surgical instrument onto the patient's three-dimensional anatomical model, align it with the key anatomical landmark points, and dynamically adjust the spatial relationship between the surgical instrument and the target position through the results of dynamic elastic compensation and multi-modal path matching for preoperative human intelligent positioning.

[0108] The method of dynamic elastic compensation includes:

[0109] Obtain the intraoperative real-time image data of the patient, and extract the dynamic deformation characteristics of the tissue in combination with the intraoperative real-time image data; construct an elastic deformation field based on a preset biomechanical model, and the preset biomechanical model includes the elastic modulus and deformation constraint conditions of different tissues;

[0110] Generate a dynamic displacement vector field according to the elastic deformation field, and map it into the patient's three-dimensional anatomical structure model in real time to correct the deformation error of the three-dimensional anatomical structure model caused by physiological movement; based on the corrected three-dimensional anatomical structure model and the high-precision optimized surgical path, dynamically adjust the position and running trajectory of the surgical instrument to align the surgical instrument with the target position.

[0111] The preset root mean square error value threshold is set by relevant staff. By collecting different root mean square error values, take the average of multiple root mean square error values as the preset root mean square error value threshold; similarly, set the preset gray value error threshold, voxel position point quantity threshold, preset motion stability threshold, and preset calcified CT threshold.

[0112] In this embodiment, by independently modeling the patient's breathing process into two stages of inhalation and exhalation, the asymmetric breathing movement of the patient can be more accurately simulated; this segmented modeling can not only improve the accurate positioning of key phase points such as the end of inhalation and the end of exhalation, but also avoid the phase shift error brought by the traditional sine wave model during the acquisition process; by setting the sampling frequency of the image acquisition device to an integer multiple of the patient's breathing frequency and performing multi-modal image acquisition at the same breathing phase, ensure that CT images, MRI images, and ultrasound images are acquired at the same physiological moment, thereby eliminating the time misalignment between different image modalities;

[0113] By setting the dynamic calcification threshold based on the statistical characteristics of the soft tissue region, it can adapt to the tissue density differences of different individuals, thus effectively avoiding misjudgment. It can be flexibly adjusted according to the specific physiological conditions of the patient, the scanning equipment, and different imaging conditions, improving the accuracy of calcification region recognition. In particular, the recognition ability for low-density calcification or calcification with blurred boundaries is greatly enhanced. By introducing the neighborhood standard deviation and combining the statistical comparison between the local region and the adjacent region, it can improve the stability and accuracy of calcification recognition when the image noise is large or the tissue boundary is blurred. It has a strong recognition ability for calcification regions with blurred boundaries, effectively avoiding the false detection problems easily occurring in traditional methods. By dynamically adjusting the offset coefficient, it can adapt to the soft tissue density fluctuations of different patients and optimize the judgment criteria for calcification regions;

[0114] By accurately marking the calcified layer as the no-go area and the large blood vessel area as the core area, and generating the minimum cost path that avoids the no-go area and the core area, it ensures that the surgical path avoids dangerous areas, improving the safety and feasibility of the surgery. The optimized surgical path not only avoids possible calcified areas but also effectively avoids large blood vessel areas, thus reducing the surgical risk. Finally, this full-process optimization method from image acquisition to preoperative intelligent positioning not only significantly improves the accuracy of surgical planning but also enhances the system's adaptability to complex morphological changes and dynamic physiological conditions, meeting the requirements of modern surgical operations for sub-millimeter-level precise positioning, and greatly improving the success rate of the surgery and the treatment effect of the patient.

[0115] Example 2

[0116] Please refer to Figure 2 As shown, for the parts not described in detail in this example, refer to the description content of Example 1. A preoperative human body intelligent positioning method for medical images is provided, including:

[0117] S1. Set the sampling frequency based on the preset respiratory gating signal, collect the multi-modal image data of the patient according to the sampling frequency, and use the point matching method to process and obtain the standardized multi-modal image data;

[0118] S2. Perform difference operations and square the gray values of different position points in the standardized multi-modal image data to obtain different squared errors; combine all the squared errors to obtain the squared error matrix, generate the error distribution heat map based on the squared error matrix; and infer tissue deformation through the error distribution heat map, and update the tissue deformation field parameters using the optimization algorithm;

[0119] S3. Based on the standardized multi-modal image data and the updated tissue deformation field parameters, generate the tissue layer mask, and assign the preset basic attenuation coefficient to construct the multi-modal attenuation rule table;

[0120] S4. Perform surgical path planning based on the tissue layer mask and the multi-modal attenuation rule table, respectively identify the positions of the calcified layer and the large blood vessels and avoid surgery to obtain an optimized surgical path;

[0121] S5. Generate an AR navigation interface by combining the optimized surgical path and the preset instrument space coordinate mapping table; based on the AR navigation interface, perform preoperative human body intelligent positioning through dynamic elastic compensation and multi-modal path matching.

[0122] Since the electronic device introduced in this embodiment is the electronic device used in the preoperative human body intelligent positioning system based on a medical image in the embodiments of the present application, based on the preoperative human body intelligent positioning system based on a medical image introduced in the embodiments of the present application, those skilled in the art can understand the specific implementation manners and various variations of the electronic device in this embodiment. Therefore, the specific implementation of how this electronic device implements the method in the embodiments of the present application will not be described in detail here. As long as those skilled in the art implement the electronic device used in the preoperative human body intelligent positioning system based on a medical image in the embodiments of the present application, it falls within the scope of protection of the present application.

[0123] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to get a formula that is closest to the actual situation. The preset parameters and threshold selection in the formulas are set by those skilled in the art according to the actual situation.

[0124] The above are only the preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the idea of the present invention belong to the protection scope of the present invention. It should be pointed out that for those ordinary technical users in the technical field, several improvements and refinements made without departing from the principle of the present invention should also be regarded as within the protection scope of the present invention.

Claims

1. A preoperative human body intelligent positioning system for medical imaging, characterized in that: include: A data standardization unit sets a sampling frequency based on a preset respiratory gating signal, collects multimodal imaging data of the patient according to the sampling frequency, and processes the data using a point matching method to obtain standardized multimodal imaging data; The deformation tracking unit performs difference calculation on the grayscale values ​​of different positions in the standardized multimodal image data and squares them to obtain different error squares; all error squares are combined to obtain an error square matrix, and an error distribution heat map is generated based on the error square matrix; The tissue deformation is inferred through the error distribution thermogram, and the parameters of the tissue deformation field are updated using an optimization algorithm; An attenuation rule construction unit generates a tissue layering mask based on the standardized multimodal imaging data and the updated tissue deformation field parameters, and assigns a preset basic attenuation coefficient to construct a multimodal attenuation rule table; The risk avoidance unit performs surgical path planning based on the tissue layering mask and the multimodal attenuation rule table, identifies the location of the calcification layer and the location of the large blood vessels and performs surgical avoidance to obtain an optimized surgical path; The navigation and positioning unit generates an AR navigation interface by combining the optimized surgical path and the preset instrument space coordinate mapping table. Based on the AR navigation interface, intelligent human body positioning before surgery is performed through dynamic elastic compensation and multi-modal path matching.

2. The preoperative human intelligent positioning system for medical imaging according to claim 1, characterized in that: The multimodal image data includes CT images, MRI images and ultrasound images.

3. The preoperative human intelligent positioning system for medical imaging according to claim 2, characterized in that: The method for acquiring standardized multimodal image data comprises: Determine the sampling frequency of the unified image acquisition device based on the respiratory gating signal, collect the patient's real-time respiratory signal, and select the end of inspiration as the trigger reference phase for unified sampling; construct a respiratory gating signal model based on the patient's real-time respiratory signal, and calibrate the key phase points in the respiratory cycle; In two consecutive respiratory cycles, the image acquisition device is triggered to acquire CT images, MRI images and ultrasound images at the same respiratory phase; the sampling frequency of the image acquisition device is set to an integer multiple of the patient's respiratory frequency, and the acquisition time point of the image acquisition device is aligned with the respiratory phase; Taking CT images as the registration benchmark, m marker points in CT images are extracted to construct a CT reference coordinate system. The corresponding marker point sets in MRI images and ultrasound images are identified using the SURF feature matching method. Different marker points in MRI images and ultrasound images are aligned with corresponding marker points in the CT reference coordinate system through an iterative nearest point algorithm to obtain the registered multimodal image data. Performing resolution uniform processing on the multimodal image data, presetting a uniform target resolution, using bilinear interpolation for the ultrasound image in the multimodal image data, and using trilinear interpolation for resampling the CT image and MRI image in the multimodal image data to achieve the preset uniform target resolution, thereby obtaining the multimodal image data after the uniform target resolution; The grayscale values ​​of the multimodal image data with unified resolution are calculated to obtain the minimum and maximum grayscale values ​​of each pixel in the CT images, MRI images and ultrasound images. The minimum and maximum grayscale values ​​of each pixel are linearly normalized to obtain the normalized multimodal image data. The grayscale value distribution of each image in the normalized multimodal image data is redistributed through the histogram equalization method to obtain the standardized multimodal image data.

4. The preoperative human intelligent positioning system for medical imaging according to claim 3, characterized in that: The method for obtaining the error distribution heat map includes: Extract CT images and MRI images from the standardized multimodal image data, extract the corresponding grayscale value for each voxel position point in the CT image and the MRI image, perform subtraction and square on the grayscale values ​​to obtain different error squares, and combine the error squares at all voxel positions in the CT image and the MRI image into an error square matrix; sum the error squares of all voxel positions, and then divide them by the total number of voxel positions to obtain the average error square, and square the average error square to obtain the root mean square error value; A root mean square error value threshold is preset. If the root mean square error value is greater than or equal to the error value threshold, the grayscale value difference between the CT image and the MRI image is judged to be abnormal and is directly marked; if the root mean square error value is less than the preset root mean square error value threshold, the grayscale value difference between the CT image and the MRI image is judged to be normal, and the error distribution heat map is generated; the error square matrix is ​​sliced ​​by layer, each layer represents an error heat map in one direction, and the grayscale value error range of each layer is normalized so that the grayscale value error is mapped to the color space. Different grayscale value error sizes will correspond to different colors, and the error heat maps of all slices are combined one by one to form an error distribution heat map.

5. The preoperative human intelligent positioning system for medical imaging according to claim 4, characterized in that: The method for updating elastic deformation field parameters comprises: A gray value error threshold and a voxel position point quantity threshold are preset. If the gray value error of the voxel position points in any area of ​​the error distribution heat map is greater than or equal to the preset gray value error threshold and greater than the preset voxel position point quantity threshold, the area is determined to be a high-risk area. If the gray value error of the voxel position points in any area of ​​the error distribution heat map is less than the preset gray value error threshold and greater than the preset voxel position point quantity threshold, the area is determined to be a low-risk area. In the error distribution heat map, the high-risk area is represented by red, and the low-risk area is represented by green. For high-risk areas, the corresponding image areas in the standardized multimodal image data are extracted, and the feature points are matched between continuous time frames using the SIFT feature matching algorithm to obtain matching results. Based on the matching results, the tissue displacement in the high-risk area is calculated. The respiratory phases are divided according to the respiratory gating signal model. In one respiratory phase, the end of exhalation is taken as the reference phase, and the tissue displacement of the inhalation phase relative to the reference phase is calculated to form a dynamic displacement table for each area. The dynamic offsets of all areas are summarized with the tissue displacements in the high-risk areas to generate a tissue deformation displacement table. According to the tissue deformation displacement table, the tissue deformation field parameters are updated through the gradient descent optimization algorithm to obtain the updated tissue deformation field parameters.

6. The preoperative human intelligent positioning system for medical imaging according to claim 5, characterized in that: The method for constructing a multi-modal attenuation rule table comprises: For the tissue deformation field of each frame of CT images and MRI images, the average displacement of all voxels is calculated as the motion intensity index of the frame to obtain the motion intensity of each frame; a motion stability threshold is preset, the motion intensity of all frames is compared with the preset motion stability threshold, and all frames with motion intensity less than the preset motion stability threshold are screened out as reference frame sets; all reference frame sets are uniformly mapped to the coordinate system of the frame through non-rigid transformation to form a deformation compensation reference coordinate system; based on the updated tissue deformation field parameters, spatial coordinate mapping is performed on the CT images and MRI images in the standardized multimodal image data set to be unified to the deformation compensation reference coordinate system; An image segmentation algorithm is used to segment the deformation-compensated CT images and MRI images and identify the target organs, generating a three-dimensional mask corresponding to each target organ; based on the updated tissue deformation field parameters, the tissue structures within the three-dimensional mask area of ​​different target organs are divided, and according to the image feature parameters and the preset tissue feature template, different tissue types of the target organs are identified, and corresponding tissue stratification masks are constructed; the image feature parameters include CT value, MRI signal intensity and T2 relaxation time difference; the position and shape of the tissue stratification mask are adjusted through the B-spline elastic deformation field to generate a fine tissue stratification mask; Based on the different tissue types of the identified target organs, the corresponding preset basic attenuation coefficients are assigned, and the tissue classification thresholds are set in combination with the image feature parameters. The basic attenuation coefficients corresponding to all tissue types of the target organs are summarized to form a basic attenuation coefficient table, and a list of basic attenuation coefficients corresponding to each tissue type is obtained; the basic attenuation coefficients of all tissue types are combined with the fine tissue stratification mask to form a multimodal attenuation rule table.

7. The preoperative human intelligent positioning system for medical imaging according to claim 6, characterized in that: The method for obtaining the optimized surgical path includes: Extract the CT grayscale value CT(x,y,z) and the corresponding tissue label La(x,y,z) at each voxel position in the tissue layer mask; the preset calcification CT threshold is T ca , define the mask function of the calcification layer, and judge whether the voxel point at the identified voxel position (x, y, z) belongs to the calcification layer area through the mask function of the calcification layer; The dynamic calcification threshold is determined by the threshold adjustment function, and the offset coefficient in the threshold adjustment function is dynamically adjusted by the offset coefficient adjustment function; a large vessel anatomical region template is preset, and based on the preset large vessel anatomical region template and the fine tissue layering mask, the position and shape of the area to be identified are matched to identify the large vessel area; and the calcification layer area is marked as a prohibited area, and the large vessel area is marked as a core area; The surgical target point and the surgical starting point are set to generate a preliminary surgical path, and the preliminary surgical path is required to avoid the restricted area and the core area. The minimum cost path to avoid the restricted area and the core area is calculated by the Dijkstra algorithm to obtain the optimized surgical path.

8. The preoperative human intelligent positioning system for medical imaging according to claim 7, characterized in that: The method for generating an AR navigation interface includes: A preset instrument space coordinate mapping table is used to perform tissue segmentation and standardization on the target organ based on standardized multimodal imaging data. Different tissue types of the target organ are distinguished by different colors or transparencies. The Marching Cubes algorithm is used to generate a three-dimensional anatomical structure model of the patient. The optimized surgical path is superimposed on the patient's three-dimensional anatomical structure model in the form of lines or arrows. According to the preset instrument space coordinate mapping table, the position and direction of the surgical instrument in the three-dimensional model are updated and displayed in real time, and an AR navigation interface is generated.

9. The preoperative human intelligent positioning system for medical imaging according to claim 8, characterized in that: The method for performing preoperative intelligent human body positioning comprises: According to the updated tissue deformation field parameters, the dynamic deformation information of the tissue at different time points is extracted, the patient's three-dimensional anatomical model and the optimized surgical path are adjusted in real time, and the respiratory gating signal model is combined with the updated tissue deformation field parameters; based on CT images, MRI images and ultrasound images, a surgical path matching table is generated, and the surgical path matching table is used to compare whether the current optimized surgical path deviates from the path in the surgical path matching table. If a deviation occurs, the current optimized surgical path is readjusted according to the surgical path matching table to obtain a high-precision optimized surgical path; Based on the three-dimensional anatomical model, key anatomical landmarks are marked, the initial position of the surgical instrument is obtained, and the initial position of the surgical instrument is mapped to the patient's three-dimensional anatomical model and aligned with the key anatomical landmarks. Through the results of dynamic elastic compensation and multimodal path matching, the spatial relationship between the surgical instrument and the target position is dynamically adjusted to perform intelligent human body positioning before surgery.

10. The medical imaging preoperative human body intelligent positioning system according to claim 9, characterized in that: The method of dynamic elastic compensation comprises: Acquire the patient's real-time intraoperative imaging data, and extract the dynamic deformation characteristics of the tissue based on the real-time intraoperative imaging data; construct an elastic deformation field based on a preset biomechanical model, which includes the elastic modulus and deformation constraints of different tissues; A dynamic displacement vector field is generated based on the elastic deformation field and mapped to the patient's three-dimensional anatomical model in real time to correct the deformation error of the three-dimensional anatomical model caused by physiological movement; based on the corrected three-dimensional anatomical model and high-precision optimized surgical path, the position and trajectory of the surgical instrument are dynamically adjusted to align the surgical instrument with the target position.

Images