Spine pedicle screw implantation system based on multi-modal image and MR navigation

Abstract

The invention provides a spinal pedicle screw implantation system based on a multi-modal image and MR navigation, and relates to the technical field of screw implantation, the system comprises a preoperative image processing module used for identifying bony characteristics of a patient and delimiting a screw implantation safety boundary according to the bony characteristics; and the bone image fusion navigation module is used for analyzing to obtain anatomical data of the patient, and calculating a screw implantation path according to a crown porcupine optimization algorithm in combination with bony characteristics. And the body position correction module is used for calculating the body position deviation value and synchronously adjusting the screw implantation path according to the body position deviation value and the screw implantation safety boundary. And the path risk early warning module is used for calculating the distance between the surgical tool and the key tissue in real time, and performing collision risk assessment and early warning. And the postoperative evaluation module is used for quantitatively evaluating the screw implantation precision and generating a postoperative report. According to the method, more accurate positioning can be provided for screw implantation, the accuracy of a screw implantation path is ensured, early warning is performed through extraction, a doctor is reminded to take measures in time, and operation safety is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of screw implantation, in particular to a spinal pedicle screw implantation system based on multi-modal imaging and MR navigation. BACKGROUND

[0002] Spinal pedicle screw implantation is an important means of treating spinal diseases, and its precision directly affects the surgical efficacy and patient prognosis. However, traditional screw implantation relies on two-dimensional imaging and surgeon experience, making it difficult to accurately identify the complex spatial relationship between bone structure and surrounding nerves and blood vessels, resulting in a high incidence of complications such as screw misplacement and nerve and blood vessel injury. On the other hand, the patient's body position during surgery is easily affected by factors such as anesthesia and surgical operation, and the existing navigation system lacks the ability to monitor and dynamically correct body position changes, further exacerbating the deviation of the implantation path. In addition, the traditional navigation method has problems such as limited two-dimensional imaging information, difficulty in multi-modal data fusion, and complex operation interface interaction, making it difficult for doctors to balance the multiple demands of safety, fixation strength, and surgical operation convenience when planning the implantation path. It is difficult to switch between image perspectives repeatedly, and the navigation operation is difficult and inefficient.

[0003] In view of the above problems, it is urgent to develop a spinal pedicle screw implantation system based on multi-modal imaging and MR navigation to improve the precision and safety of screw implantation, reduce the difficulty of intraoperative operation, and improve the surgical prognosis. SUMMARY

[0004] The present application provides a spinal pedicle screw implantation system based on multi-modal imaging and MR navigation to solve the problem of insufficient precision of screw implantation in the prior art, the influence of intraoperative body position changes on implantation accuracy, and the high difficulty of navigation operation of the implantation path.

[0005] The present application provides a spinal pedicle screw implantation system based on multi-modal imaging and MR navigation, comprising: A preoperative image processing module is used to collect the patient's preoperative bone structure image and nerve and blood vessel soft tissue image, identify the patient's bone features using a deep learning segmentation model, and define a screw implantation safety boundary based on the bone features.

[0006] A bone image fusion navigation module is used to fuse and analyze the bone structure image and nerve and blood vessel soft tissue image to obtain patient anatomical data according to the image parameterization positioning method, and calculate the screw implantation path based on the bone features according to the crown hog optimization algorithm.

[0007] A body position correction module is used to display the bone features and intraoperative patient bone structure superimposed through a mixed reality glasses to calculate the body position deviation, and adjust the screw implantation path in synchronization with the body position deviation and the screw implantation safety boundary through gestures.

[0008] The path risk warning module is used to calculate the distance between the surgical tool and the bony safety boundary and neurovascular tissue in the screw implantation path in real time based on the distance field algorithm and the near-distance penetration prediction algorithm, and to conduct collision risk assessment and warning.

[0009] The postoperative assessment module is used to quantify the accuracy of screw implantation and generate a postoperative report based on bony characteristics.

[0010] This invention provides a spinal pedicle screw implantation system based on multimodal imaging and MR navigation. The steps of obtaining bony features in the preoperative image processing module include: The images of skeletal structures and neurovascular soft tissues were converted into the format of a deep learning segmentation model, and then spatial resampling, denoising, and region extraction were performed to obtain preprocessed image data.

[0011] A multimodal fusion segmentation model is used as a deep learning segmentation model. It outputs preprocessed image data, collaboratively learns the feature associations between images, and segments bones and soft tissues to obtain bony structure masks.

[0012] A three-dimensional geometric feature detection algorithm is used to locate the bony landmarks of the bony structure mask, and the bony features are obtained by converting the pixel spacing into physical coordinates.

[0013] This invention provides a spinal pedicle screw implantation system based on multimodal imaging and MR navigation. The steps of obtaining the screw implantation safety boundary by the preoperative image processing module include: Based on the bony features, the constraint factors affecting screw implantation safety are identified. According to the spatial coordinate system of the bone structure image and the neurovascular soft tissue image, the three-dimensional coordinates of the bony features are extracted to construct a three-dimensional bony model.

[0014] The polyhedral boundary modeling algorithm is used to transform the constraint elements into allowed and prohibited areas in three-dimensional space, forming a three-dimensional safety boundary model.

[0015] By overlaying the 3D bony model with the 3D safety boundary model, and using color coding to distinguish between safe and dangerous areas, the screw implantation safety boundary is obtained.

[0016] This invention provides a spinal pedicle screw implantation system based on multimodal imaging and MR navigation. The steps of obtaining patient anatomical data by the bone image fusion navigation module include: The three-dimensional coordinates of bony features are extracted from skeletal images, and the spatial transformation parameters of neurovascular soft tissue images relative to skeletal images are calculated using the nearest point iteration algorithm.

[0017] Based on spatial transformation parameters, the neurovascular soft tissue images are rigidly adjusted so that the bony features of the two types of images completely overlap to obtain a fused image.

[0018] The fused images are input into a deep learning segmentation model, and the skeletal geometric features and soft tissue semantic features are extracted as fusion features to output a mask of skeletal and soft tissue structures.

[0019] Based on the masking of skeletal and soft tissue structures, combined with the three-dimensional coordinates of bony features, the bony parameters and soft tissue correlation parameters are calculated as patient anatomical data through geometric analysis and medical algorithms.

[0020] This invention provides a spinal pedicle screw implantation system based on multimodal imaging and MR navigation. The steps of obtaining the screw implantation path by the bone image fusion navigation module include: With safety, fixation strength, and surgical operation boundaries as objectives, a multi-objective optimization function is constructed, and constraints are set from the needle entry point, angle, and length.

[0021] The biological characteristics of the hog optimization algorithm are correlated with path optimization. An initial population is generated based on skeletal features, with each hog corresponding to a candidate path.

[0022] For each individual crowned porcupine, the direction of the spines is determined based on the bony characteristics and the current optimal solution, and a preset number of spines are generated.

[0023] The step length is determined based on the pedicle diameter, and the direction of the thorns is combined to adjust each individual crowned porcupine to generate a new individual.

[0024] The initial population is merged with the newly generated individuals. The objective function value of each individual is calculated and sorted. A new population is formed by selecting a preset number of individuals.

[0025] After the maximum number of iterations is reached, the individual with the largest objective function value is selected as the screw implantation path.

[0026] This invention provides a spinal pedicle screw implantation system based on multimodal imaging and MR navigation. The step of obtaining the postural deviation by the postural correction module includes: Establish an initial mapping relationship between the skeletal 3D model and the device coordinate system of the mixed reality glasses, and generate an initial transformation matrix.

[0027] Markers are fixed in the surgical area of ​​the patient at a position that is relatively static relative to the bony structures of the spine. A real-time coordinate system is generated in the surgical field using mixed reality glasses, and the real-time positional changes of the markers are calculated as a reference for body position tracking.

[0028] Based on the body position tracking benchmark, the initial transformation matrix is ​​optimized to generate a temporary calibration matrix, which is then superimposed onto the patient's actual surgical field during the operation to generate virtual and real bony feature coordinates.

[0029] The translation deviation is obtained by calculating the average translation deviation of the bony features based on the coordinates of the virtual and real bony features using the least squares method.

[0030] Based on the virtual and real bony feature coordinates, the patient's rotational deviation during surgery relative to the preoperative state is calculated using the singular value decomposition algorithm.

[0031] Translational and rotational deviations are displayed in real time as positional deviations using mixed reality glasses.

[0032] This invention provides a spinal pedicle screw implantation system based on multimodal imaging and MR navigation. The step of adjusting the screw implantation path by the body position correction module includes: By binding gesture commands to path adjustment actions, a mapping adjustment relationship is obtained, and key constraints are set based on the body position deviation and screw implantation safety boundary.

[0033] Based on the positional deviation displayed in the mixed reality glasses, the screw implantation path is corrected as a whole through mapping adjustment relationships, and fine adjustments are made according to key constraints.

[0034] This invention provides a path risk warning module for a spinal pedicle screw implantation system based on multimodal imaging and MR navigation, comprising: The safe screw implantation prediction unit is used to construct a distance field model based on the screw implantation safety boundary, and to collect historical surgical data and combine it with a random forest classifier to construct a near-range penetration prediction model.

[0035] The intraoperative safety distance measurement unit is used to acquire the three-dimensional coordinates of surgical tools in real time during the operation, and to calculate the bony safety distance and neurovascular distance by combining the distance field model.

[0036] The safety boundary classification unit is used to classify the risk level of bony safety distance and neurovascular distance based on the screw implantation safety boundary.

[0037] The risk assessment and grading unit is used to predict the risk of surgical tool trajectory based on the close-range penetration prediction model to obtain multiple predicted risk levels, and to integrate the multiple risk levels to obtain the final assessment result.

[0038] The early warning and response unit is used to trigger early warnings through mixed reality glasses, sound, and touch based on the final assessment results, and simultaneously display risk tracing information and operational suggestions.

[0039] This invention provides a spinal pedicle screw implantation system based on multimodal imaging and MR navigation. The steps of constructing a distance field model and a near-field penetration prediction model by the safe screw implantation prediction unit include: The shortest Euclidean distance from each voxel in the computational space to the screw implantation safety boundary is calculated to generate distance field data, and a distance field model is constructed in accordance with the intraoperative image voxels.

[0040] A training dataset is formed by labeling distance field data and actual penetration results from historical surgical data. A prediction model is built using a random forest classifier, with the current tool position, movement speed and type as input features and the penetration risk level as the output label to obtain a near-range penetration prediction model.

[0041] This invention provides a spinal pedicle screw implantation system based on multimodal imaging and MR navigation. The postoperative evaluation module generates a postoperative report, including the following steps: After the screw implantation surgery is completed, three-dimensional image data of the surgical segment of the patient is acquired, and a postoperative bony feature reference set is generated based on the corresponding anatomical location of the bony feature markers.

[0042] Using the postoperative bony feature benchmark set as a reference, the planned path parameters are obtained by aligning the preoperative implantation path with the postoperative image coordinate system through an iterative nearest point algorithm.

[0043] Using the planned path parameters as a reference, the screw implantation accuracy is classified into accuracy levels based on the deviation of the needle insertion point, angle deviation, depth deviation, and cortical penetration, and a postoperative report containing a three-dimensional comparison map of the path, evaluation conclusions, and improvement suggestions is generated.

[0044] This invention provides a spinal pedicle screw implantation system based on multimodal imaging and MR navigation. Through multimodal image fusion and a deep learning segmentation model, it can more accurately identify bony features, delineate safe boundaries for screw implantation, and provide more precise positioning for screw implantation, reducing damage to surrounding nerve and blood vessel tissues. Real-time monitoring and correction of intraoperative positioning deviations ensure the accuracy of the screw implantation path, avoiding surgical errors caused by changes in body position and reducing surgical risks. Real-time calculation of the distance between surgical tools and bony safety boundaries and nerve and blood vessel tissues accurately assesses collision risks and provides early warnings through multiple methods, reminding doctors to take timely measures to ensure surgical safety. Early warnings are provided through mixed reality glasses, sound, and touch, enabling doctors to perceive surgical risks more promptly and comprehensively, improving the efficiency of risk management. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0046] Fig. 1 This is one of the flowcharts of the spinal pedicle screw implantation system based on multimodal imaging and MR navigation provided in the embodiments of the present invention; Fig. 2This is the second schematic diagram of the process of the spinal pedicle screw implantation system based on multimodal imaging and MR navigation provided in the embodiments of the present invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0048] The following is combined Figs. 1-2 This invention describes a spinal pedicle screw implantation system based on multimodal imaging and MR navigation.

[0049] like Fig. 1 As shown, the spinal pedicle screw implantation system based on multimodal imaging and MR navigation provided in this embodiment of the invention includes: The preoperative image processing module is used to collect preoperative skeletal structure images and neurovascular soft tissue images of patients, and combine them with a deep learning segmentation model to identify the patient's bony features and delineate the safe boundary for screw implantation based on the bony features.

[0050] Skeletal structure imaging acquisition: 64-slice or higher spiral CT (such as Siemens SOMATOM Definition Flash) is used, whose high spatial resolution can clearly show the cortical / cancellous bone boundary of bony structures such as vertebral bodies, pedicles, transverse processes, and articular processes.

[0051] Scanning parameter settings: Tube voltage is used to balance bone clarity and radiation dose, and can be set to 120-140kV. Tube current: Adjusted according to patient weight to avoid image noise, and can be set to 200-300mA.

[0052] The target surgical segment and one vertebra above and below it are covered. Bone reconstruction is performed using a bone algorithm to generate three-dimensional volumetric data, preserving detailed features of the bony structure, such as pedicle diameter and zigzag ridge morphology.

[0053] Neurovascular soft tissue imaging: 1.5T or higher superconducting MRI (such as GESignaHDxt) was used to distinguish the signal differences between soft tissues such as nerves, blood vessels, and spinal cord and surrounding tissues through different sequences.

[0054] Key sequence selection and parameters may include: T1-weighted images (T1WI): used to locate the anatomical relationship between the vertebral body and soft tissues, clearly showing the spatial location of the vertebral body margins, pedicles, and spinal cord. T2-weighted images (T2WI): highlight neurovascular signals to identify the course of nerves and their adjacent relationship to the pedicles. Enhanced MRI (e.g., Gd-DTPA enhancement): if the patient has spinal tumors, infections, or other lesions, contrast agents are needed to enhance the signals of blood vessels and diseased tissues, eliminating interference from the lesions on the identification of bony features.

[0055] The steps for obtaining bony features in the preoperative image processing module include: The images of skeletal structures and neurovascular soft tissues were converted into the format of a deep learning segmentation model, and then spatial resampling, denoising, and region extraction were performed to obtain preprocessed image data.

[0056] Spatial resampling includes: uniformly adjusting the voxel size of skeletal structure images (CT) and neurovascular soft tissue images (MRI) to 1×1×1mm using a linear interpolation algorithm to ensure that the coordinates of the same anatomical location are completely corresponding in the two types of images, laying the foundation for multimodal fusion.

[0057] Denoising processing includes: CT image denoising: Gaussian filtering is used to smooth image noise, while adaptive thresholding is used to preserve the edge features of bony structures. If metal artifacts are present, the Metal Artifact Correction (MAR) algorithm is used to eliminate the interference of artifact regions on bone segmentation.

[0058] MRI image denoising: To address the high noise characteristics of T2WI sequences, nonlocal mean filtering is used to reduce noise. At the same time, motion artifact correction algorithms are used to repair image blurring caused by the patient's breathing and heartbeat, ensuring clear boundaries of soft tissues such as nerve roots and spinal cord.

[0059] Region extraction may include: automatically cropping irrelevant anatomical regions from CT and MRI images based on surgeon-annotated surgical segments using region growing algorithms, reducing the interference of redundant data on model training / inference. Gray-level normalization is then performed on the cropped regions: mapping the CT values ​​of CT images to the range of 0-255, and standardizing the signal intensity of each MRI sequence to the same gray-level range, ensuring the model has consistent recognition capabilities for images from different patients and from different devices.

[0060] A multimodal fusion segmentation model is used as a deep learning segmentation model. It outputs preprocessed image data, collaboratively learns the feature associations between images, and segments bones and soft tissues to obtain bony structure masks.

[0061] Employing the ViT-UNet (a combination of Visual Transformer and U-Net) architecture, it can simultaneously capture details of bony structures and extensive features of soft tissues. A labeled dataset containing over 500 preprocessed spinal images was constructed. Target structures, including vertebral bodies, pedicles, transverse processes, superior / inferior articular processes, spinal cord, nerve roots, and vertebral arteries, were manually labeled using annotation tools by at least two chief physicians of spinal surgery, forming segmentation masks.

[0062] The Dice similarity coefficient (DSC) was used as the primary evaluation metric, and the Adam optimizer was employed with an initial learning rate of 1e-4. Early stopping was used to prevent overfitting. During training, a multimodal feature attention mechanism was used: the model automatically focused on bony high-signal areas in CT and soft tissue signal difference areas in MRI, strengthening the fusion of complementary information from the two types of images and avoiding segmentation errors caused by the lack of information from a single modality.

[0063] The preprocessed image data is input into the trained deep learning segmentation model, and the probability mask of each target structure is output through forward propagation.

[0064] An adaptive thresholding method is used to convert the probability mask into a binary segmentation mask, which clearly distinguishes the spatial range between bony structures such as vertebral bodies and pedicles and soft tissues such as spinal cord and nerve roots.

[0065] By eliminating tiny holes in the segmentation mask through morphological closing operations and removing isolated noise regions through connected component analysis, the integrity of the skeletal structure is ensured to obtain the skeletal structure mask.

[0066] A three-dimensional geometric feature detection algorithm is used to locate the bony landmarks of the bony structure mask, and the bony features are obtained by converting the pixel spacing into physical coordinates.

[0067] Bony landmarks may include: pedicle entry point: By calculating the intersection of the upper and lateral edges of the pedicle, a reference starting point for screw insertion is automatically marked to determine the screw diameter and length.

[0068] Vertex apex: Identifies the highest point of the bony ridge between the mastoid and accessory processes in a lumbar segmentation mask. Used to assess vertebral body load-bearing capacity and select screw insertion depth.

[0069] Midline / superior border of transverse process: The tangent line between the three-dimensional center point of the transverse process and the apex of the transverse process is used as a reference line for needle insertion angle calibration. It helps determine the lateral position of the needle insertion point and avoids damage to the transverse nerve.

[0070] Facet joint space: By analyzing the grayscale difference between the lower edge of the superior facet joint and the upper edge of the inferior facet joint, the position of the joint space is located to avoid screw penetration of the articular surface. The lateral safety boundary for screw insertion is calibrated.

[0071] The steps for obtaining the screw implantation safety boundary using the preoperative image processing module include: Based on the bony features, the constraint factors affecting screw implantation safety are identified. According to the spatial coordinate system of the bone structure image and the neurovascular soft tissue image, the three-dimensional coordinates of the bony features are extracted to construct a three-dimensional bony model.

[0072] The polyhedral boundary modeling algorithm is used to transform the constraint elements into allowed and prohibited areas in three-dimensional space, forming a three-dimensional safety boundary model.

[0073] By overlaying the 3D bony model with the 3D safety boundary model, and using color coding to distinguish between safe and dangerous areas, the screw implantation safety boundary is obtained.

[0074] The bone image fusion navigation module is used to fuse and analyze skeletal structure images and neurovascular soft tissue images according to the image parametric localization method to obtain patient anatomical data, and calculate the screw implantation path based on the porcupine optimization algorithm in combination with bone features.

[0075] The steps by which the bone fusion navigation module obtains patient anatomical data include: The three-dimensional coordinates of bony features are extracted from skeletal images, and the spatial transformation parameters of neurovascular soft tissue images relative to skeletal images are calculated using the nearest point iteration algorithm.

[0076] Based on spatial transformation parameters, the neurovascular soft tissue images are rigidly adjusted so that the bony features of the two types of images completely overlap to obtain a fused image.

[0077] The fused images are input into a deep learning segmentation model, and the skeletal geometric features and soft tissue semantic features are extracted as fusion features to output a mask of skeletal and soft tissue structures.

[0078] Based on the masking of skeletal and soft tissue structures, combined with the three-dimensional coordinates of bony features, the bony parameters and soft tissue correlation parameters are calculated as patient anatomical data through geometric analysis and medical algorithms.

[0079] The steps for calculating bony parameters may include: Pedicle diameter: On the maximum cross-section of the pedicle, extract the cortical boundary pixel coordinates of the bone and soft tissue structure mask, calculate the left and right diameters and the front and back diameters using the "minimum bounding rectangle algorithm", and convert them into the actual physical diameter by combining the pixel spacing.

[0080] Pedicle angle: Extract the central axis of the pedicle, calculate the inclination angle between the axis and the coronal plane of the vertebral body, and the cephalic / coccygeal tilt angle with respect to the sagittal plane, with the angle error controlled within ≤1°.

[0081] Bony spinal canal dimensions: The anteroposterior and lateral diameters are calculated at the maximum cross-section of the spinal canal to provide basic bony data for assessing the degree of spinal stenosis.

[0082] The steps for calculating soft tissue correlation parameters may include: degree of spinal stenosis: combining the spinal cord soft tissue mask, calculating the area occupied by the spinal cord in the spinal canal, the ratio of the area to the area of ​​the bony spinal canal, and determining the degree of bony stenosis and functional stenosis based on the size of the bony spinal canal.

[0083] Neurovascular-skeletal distance: The minimum distance between the nerve root mask and the medial cortical mask of the pedicle, and the minimum distance between the vertebral artery mask and the anterior wall mask of the pedicle are calculated using the "Euclidean distance algorithm" to provide data support for subsequent screw implantation path safety assessment.

[0084] like Fig. 2 As shown, the steps for the bone image fusion navigation module to obtain the screw implantation path include: With safety, fixation strength, and surgical operation boundaries as objectives, a multi-objective optimization function is constructed, and constraints are set from the needle entry point, angle, and length.

[0085] The biological characteristics of the hog optimization algorithm are correlated with path optimization. An initial population is generated based on skeletal features, with each hog corresponding to a candidate path.

[0086] For each individual crowned porcupine, the direction of the spines is determined based on bony characteristics and the current optimal solution, and a preset number of spines are generated, expressed by the formula:

[0087] In the formula, It is the first The first individual Root thorn direction angle, It is a basic orientation angle based on bony characteristics. It is the direction angle of the current optimal solution. , It's weight. It's a random angle.

[0088] The step length is determined based on the pedicle diameter, and the direction of the spines is considered when adjusting each individual crowned porcupine to generate a new individual. The formula is expressed as follows:

[0089] In the formula, It is a new individual. It is a crowned porcupine individual. It's the step length. It's in the direction of the thorns. It is the length adjustment amount.

[0090] The initial population is merged with the newly generated individuals. The objective function value of each individual is calculated and sorted. A new population is formed by selecting a preset number of individuals.

[0091] After the maximum number of iterations is reached, the individual with the largest objective function value is selected as the screw implantation path.

[0092] The positioning correction module is used to overlay bony features with the patient's bony structure during surgery using mixed reality glasses to calculate the positioning deviation. Based on the positioning deviation and the screw implantation safety boundary, the screw implantation path is adjusted synchronously by gestures.

[0093] The steps by which the body position correction module obtains the body position deviation include: Establish an initial mapping relationship between the skeletal 3D model and the device coordinate system of the mixed reality glasses, and generate an initial transformation matrix.

[0094] Markers are fixed in the surgical area of ​​the patient at a position that is relatively static relative to the bony structures of the spine. A real-time coordinate system is generated in the surgical field using mixed reality glasses, and the real-time positional changes of the markers are calculated as a reference for body position tracking.

[0095] Based on the body position tracking benchmark, the initial transformation matrix is ​​optimized to generate a temporary calibration matrix, which is then superimposed onto the patient's actual surgical field during the operation to generate virtual and real bony feature coordinates.

[0096] The translational deviation is calculated using the least squares method based on the coordinates of the real and virtual bony features. The formula is as follows:

[0097] In the formula, It is the translation deviation rate. It is the magnitude of the translational deviation of a single bony feature. It is the first The weight of each marker It is the number of markers.

[0098] Based on the virtual and real bony feature coordinates, the patient's rotational deviation during surgery relative to the preoperative state is calculated using the singular value decomposition algorithm.

[0099] The steps for calculating rotational deviation include: The formula for calculating the center point of the preoperative virtual landmark and the center point of the intraoperative real landmark is as follows:

[0100]

[0101] In the formula, It is the preoperative center point. It is the center point during the operation. It is the first The three-dimensional coordinates of a preoperative virtual landmark. It is the first The three-dimensional coordinates of the actual landmark points during the operation.

[0102] The decentered coordinate matrix is ​​constructed using the following formula:

[0103]

[0104] In the formula, It is the preoperative decentered coordinate matrix. It is the intraoperative decentered coordinate matrix.

[0105] For matrix Perform singular value decomposition to obtain Then the rotation matrix .in yes The transpose of the matrix, It is the left singular vector matrix obtained from singular value decomposition. It is a diagonal matrix. It is the right singular vector matrix obtained from singular value decomposition. yes The transpose of .

[0106] Converting the rotation matrix into Euler angles outputs the rotation deviation, which can be "rotating 1.5° around the Y-axis of the coronal plane, rotating 0.6° around the Z-axis of the cross section, etc."

[0107] Translational and rotational deviations are displayed in real time as positional deviations using mixed reality glasses.

[0108] The steps for adjusting the screw implantation path using the body positioning correction module include: By binding gesture commands to path adjustment actions, a mapping adjustment relationship is obtained, and key constraints are set based on the body position deviation and screw implantation safety boundary.

[0109] For example, single-finger sliding: controlling the translation of the screw path in the coronal / sagittal plane, sliding the index finger to the right = the path is translated to the right, sliding it upwards = the path is translated to the head.

[0110] Two-finger rotation: The screw implantation angle can be adjusted by rotating the thumb and index finger clockwise = increasing the coronal inclination angle, and counterclockwise = decreasing the inclination angle.

[0111] Fist clenching and unclenching: The length adjustment of the control screw can increase the length when the fist is clenched and then opened, and decrease the length when the fist is clenched and then tightened.

[0112] Double-tap your palm: Confirm the current adjustment.

[0113] Translation and rotation deviations are displayed in the form of "3D arrows + numerical values". The arrow color changes with the size of the deviation, for example, ≤1mm is green, 1-2mm is yellow, and >2mm is red.

[0114] A semi-transparent colored grid is overlaid on the actual bone structure during surgery, with green indicating safe areas, red indicating dangerous areas, and yellow indicating warning areas. A 3D model of the current screw path is also displayed, which can be shown as a blue outline, with the relative positions of the insertion point, screw tip, and safe boundary updated in real time.

[0115] Based on the positional deviation displayed in the mixed reality glasses, the screw implantation path is corrected as a whole through mapping adjustment relationships, and fine adjustments are made according to key constraints.

[0116] If the patient's positional deviation shows that he has shifted 1.0 mm to the right, the doctor will use a single finger to slide to the left and observe the blue path frame moving synchronously in the MR field of view until the starting point of the path is realigned with the pre-planned bony feature point.

[0117] If the patient's positional deviation indicates that the patient has rotated 1.2° around the coronal plane, the doctor can adjust the path angle by using a two-finger counterclockwise rotation gesture to bring the angle between the path axis and the pedicle center axis back to a safe range.

[0118] If the initial adjustment path approaches the yellow warning zone, the doctor uses a two-finger micro-rotation gesture to increase the screw's inclination angle by 0.5°, moving the path towards the center of the safe zone. If the path's endpoint approaches the safe boundary of the anterior vertebral cortex, the doctor uses a clenched fist gesture to reduce the screw length, increasing the distance between the screw tip and the anterior cortex. During adjustment, the system automatically checks whether the length meets the fixation strength requirements; if not, it guides the doctor through text prompts.

[0119] The path risk warning module is used to calculate the distance between the surgical tool and the bony safety boundary and neurovascular tissue in the screw implantation path in real time based on the distance field algorithm and the near-distance penetration prediction algorithm, and to conduct collision risk assessment and warning.

[0120] The path risk warning module includes: The safe screw implantation prediction unit is used to construct a distance field model based on the screw implantation safety boundary, and to collect historical surgical data and combine it with a random forest classifier to construct a near-range penetration prediction model.

[0121] The steps for constructing the distance field model and near-range penetration prediction model for the safety anchor pre-judgment unit include: The shortest Euclidean distance from each voxel in the computational space to the screw implantation safety boundary is calculated to generate distance field data, which is then used to construct a distance field model that is consistent with the intraoperative image voxels. The distance field model is dynamically adjusted based on real-time positional deviations: when the patient's position shifts or rotates, the spatial coordinates of each voxel are corrected using a coordinate transformation matrix to ensure that the distance field data always remains consistent with the actual intraoperative anatomical position, avoiding distance calculation errors caused by positional changes.

[0122] A training dataset is formed by labeling distance field data and actual penetration results from historical surgical data. A prediction model is built using a random forest classifier, with the current tool position, movement speed and type as input features and the penetration risk level as the output label to obtain a near-range penetration prediction model.

[0123] The intraoperative safety distance measurement unit is used to acquire the three-dimensional coordinates of surgical tools in real time during the operation, and to calculate the bony safety distance and neurovascular distance by combining the distance field model.

[0124] The steps for calculating bony safety distances and neurovascular distances may include: inputting virtual and real bony feature coordinates into a distance field model, and calculating the shortest distance from the tool tip to the nearest bony safety boundary and the nearest neurovascular tissue by addressing the problem of non-integer voxels in the tool coordinates.

[0125] The safety boundary classification unit is used to classify the risk level of bony safety distance and neurovascular distance based on the screw implantation safety boundary.

[0126] Risk Level Table:

[0127] The risk assessment and grading unit is used to predict the risk of surgical tool trajectory based on the close-range penetration prediction model to obtain multiple predicted risk levels, and to integrate the multiple risk levels to obtain the final assessment result.

[0128] The early warning and response unit is used to trigger early warnings through mixed reality glasses, sound, and touch based on the final assessment results, and simultaneously display risk tracing information and operational suggestions.

[0129] The postoperative assessment module is used to quantify the accuracy of screw implantation and generate a postoperative report based on bony characteristics.

[0130] The steps involved in generating a postoperative report using the postoperative assessment module include: After the screw implantation surgery is completed, three-dimensional image data of the surgical segment of the patient is acquired, and a postoperative bony feature reference set is generated based on the corresponding anatomical location of the bony feature markers.

[0131] Using the postoperative bony feature benchmark set as a reference, the planned path parameters are obtained by aligning the preoperative implantation path with the postoperative image coordinate system through an iterative nearest point algorithm.

[0132] Using the planned path parameters as a reference, the screw implantation accuracy is classified into accuracy levels based on the deviation of the needle insertion point, angle deviation, depth deviation, and cortical penetration, and a postoperative report containing a three-dimensional comparison map of the path, evaluation conclusions, and improvement suggestions is generated.

[0133] The spinal pedicle screw implantation system based on multimodal imaging and MR navigation provided in this embodiment accurately identifies bony features and generates screw implantation paths through a deep learning segmentation model and the Crown Porcupine optimization algorithm, significantly improving the accuracy of screw implantation. Furthermore, it can adjust the screw implantation path in a timely manner to avoid collisions between surgical tools and bony safety boundaries and neurovascular tissues, reducing surgical risks. The use of mixed reality glasses provides precise navigation and can monitor the distance between the tool and the patient's critical tissues in real time. When a collision risk is detected, the system will issue an alarm promptly, greatly improving the accuracy and safety of the surgery. Moreover, through multimodal image fusion and automated path planning, preoperative preparation time and intraoperative adjustment time are reduced, improving surgical efficiency.

[0134] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0135] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A spinal pedicle screw implantation system based on multimodal imaging and MR navigation, characterized in that, include: The preoperative image processing module is used to collect preoperative skeletal structure images and neurovascular soft tissue images of patients, combine them with a deep learning segmentation model to identify the patient's bony features, and delineate the safe boundary for screw implantation based on the bony features. The bone image fusion navigation module is used to fuse and analyze the bone structure image and the neurovascular soft tissue image according to the image parametric localization method to obtain the patient's anatomical data, and calculate the screw implantation path according to the crown porcupine optimization algorithm based on the bone features. The positioning correction module is used to overlay the bony features with the patient's bony structure during surgery using mixed reality glasses to calculate the positioning deviation, and to adjust the screw implantation path synchronously by gestures based on the positioning deviation and the screw implantation safety boundary. The path risk warning module is used to calculate the distance between the surgical tool and the bony safety boundary and neurovascular tissue in the screw implantation path in real time based on the distance field algorithm and the near-distance penetration prediction algorithm, and to conduct collision risk assessment and warning. The postoperative assessment module is used to quantify the screw implantation accuracy and generate a postoperative report based on the aforementioned bony characteristics.

2. The spinal pedicle screw implantation system based on multimodal imaging and MR navigation according to claim 1, characterized in that, The steps by which the preoperative image processing module obtains the bony features include: The skeletal structure images and the neurovascular soft tissue images are converted into the format of the deep learning segmentation model, and spatial resampling, denoising and region extraction are performed to obtain preprocessed image data. A multimodal fusion segmentation model is used as the deep learning segmentation model to output the preprocessed image data, collaboratively learn the feature associations between images, and segment bones and soft tissues to obtain a bony structure mask. A three-dimensional geometric feature detection algorithm is used to locate the bony landmarks of the bony structure mask, and the bony features are obtained by converting the pixel spacing into physical coordinates.

3. The spinal pedicle screw implantation system based on multimodal imaging and MR navigation according to claim 1, characterized in that, The steps by which the preoperative image processing module obtains the screw implantation safety boundary include: Based on the aforementioned bony features, constraint factors affecting screw implantation safety are determined. According to the spatial coordinate system of the bone structure image and the neurovascular soft tissue image, the three-dimensional coordinates of the bony features are extracted to construct a three-dimensional bony model. The constraint elements are transformed into allowed and prohibited regions in three-dimensional space by using a polyhedral boundary modeling algorithm to form a three-dimensional safety boundary model. The bone-like three-dimensional model is superimposed with the three-dimensional safety boundary model, and the safe area and dangerous area are distinguished by color coding to obtain the screw implantation safety boundary.

4. The spinal pedicle screw implantation system based on multimodal imaging and MR navigation according to claim 1, characterized in that, The steps by which the bone image fusion navigation module obtains the patient's anatomical data include: The three-dimensional coordinates of the bony features are extracted from the skeletal structure image, and the spatial transformation parameters of the neurovascular soft tissue image relative to the skeletal structure image are calculated using the nearest point iteration algorithm. The neurovascular soft tissue images are rigidly adjusted according to the spatial transformation parameters so that the bony features of the two types of images completely overlap to obtain a fused image. The fused image is input into the deep learning segmentation model, and the skeletal geometric features and soft tissue semantic features are extracted as fusion features. The skeletal and soft tissue structure mask is output. Based on the skeletal and soft tissue structure mask, and combined with the three-dimensional coordinates of the bony features, the bony parameters and soft tissue correlation parameters are calculated using geometric analysis and medical algorithms as the patient's anatomical data.

5. The spinal pedicle screw implantation system based on multimodal imaging and MR navigation according to claim 1, characterized in that, The steps by which the bone image fusion navigation module obtains the screw implantation path include: With safety, fixation strength, and surgical operation boundaries as objectives, a multi-objective optimization function was constructed, and constraints were set from the needle entry point, angle, and length. The biological characteristics of the hog optimization algorithm are correlated with path optimization. An initial population is generated based on skeletal features, with each hog corresponding to a candidate path. For each individual crowned porcupine, the direction of the spines is determined based on the aforementioned bony characteristics and the current optimal solution, and a preset number of spines are generated; The step length is determined based on the pedicle diameter, and the direction of the thorns is combined to adjust each individual of the crowned porcupine to generate a new individual; The initial population is merged with the newly generated individuals, the objective function value of each individual is calculated and sorted, and a new population is formed by selecting a preset number of individuals. After the maximum number of iterations is reached, the individual with the largest objective function value is selected as the screw implantation path.

6. The spinal pedicle screw implantation system based on multimodal imaging and MR navigation according to claim 3, characterized in that, The steps by which the body position correction module obtains the body position deviation include: Establish an initial mapping relationship between the skeletal 3D model and the device coordinate system of the mixed reality glasses, and generate an initial transformation matrix; Markers are fixed in the surgical area of ​​the patient at a position that is relatively static relative to the bony structures of the spine. A real coordinate system is generated during the operation using mixed reality glasses, and the real-time positional changes of the markers are calculated as a reference for body position tracking. Based on the body position tracking benchmark, the initial transformation matrix is ​​optimized to generate a temporary calibration matrix, which is then superimposed onto the patient's actual surgical field during the operation to generate virtual and real bony feature coordinates. The translation deviation is obtained by calculating the average translation deviation of the bony features based on the virtual and real bony feature coordinates using the least squares method. Based on the aforementioned virtual and real bony feature coordinates, the patient's intraoperative rotational deviation relative to the preoperative state is calculated using a singular value decomposition algorithm. The translational deviation and the rotational deviation are displayed in real time as the body position deviation through the mixed reality glasses.

7. The spinal pedicle screw implantation system based on multimodal imaging and MR navigation according to claim 1, characterized in that, The steps of adjusting the screw implantation path by the body positioning correction module include: The gesture commands are bound to the path adjustment actions to obtain the mapping adjustment relationship, and key constraints are set based on the body position deviation and the screw implantation safety boundary. Based on the body position deviation displayed in the mixed reality glasses, the screw implantation path is corrected as a whole through the mapping adjustment relationship, and fine-tuned according to the key constraints.

8. The spinal pedicle screw implantation system based on multimodal imaging and MR navigation according to claim 1, characterized in that, The path risk warning module includes: The safe screw implantation prediction unit is used to construct a distance field model based on the screw implantation safety boundary, and to collect historical surgical data and combine it with a random forest classifier to construct a near-range penetration prediction model. The intraoperative safety distance measurement unit is used to acquire the three-dimensional coordinates of surgical tools in real time during the operation, and calculate the bony safety distance and neurovascular distance in combination with the distance field model. A safety boundary classification unit is used to classify the risk level of the bony safety distance and the neurovascular distance based on the screw implantation safety boundary; The risk assessment and grading unit is used to predict the risk of surgical tool trajectory based on the near-distance penetration prediction model to obtain multiple predicted risk levels, and to integrate the multiple risk levels to obtain the final assessment result. The early warning and linkage response unit is used to trigger an early warning through mixed reality glasses, sound and touch based on the final assessment results, and simultaneously display risk tracing information and operation suggestions.

9. The spinal pedicle screw implantation system based on multimodal imaging and MR navigation according to claim 8, characterized in that, The steps of constructing the distance field model and the near-range penetration prediction model by the security nail implantation prediction unit include: The shortest Euclidean distance from each voxel in the computational space to the screw implantation safety boundary is calculated to generate distance field data, and a distance field model is constructed in consistency with the intraoperative image voxels. A training dataset is formed by labeling distance field data and actual penetration results from the historical surgical data. A prediction model is constructed using the random forest classifier, with the current tool position, movement speed and type as input features and the penetration risk level as the output label, to obtain a near-range penetration prediction model.

10. The spinal pedicle screw implantation system based on multimodal imaging and MR navigation according to claim 1, characterized in that, The steps for the postoperative assessment module to generate the postoperative report include: After the screw implantation surgery is completed, three-dimensional image data of the surgical segment of the patient is acquired, and a postoperative bony feature reference set is generated based on the anatomical location corresponding to the bony feature markers. Using the postoperative bony feature reference set as a reference, the planned path parameters are obtained by aligning the preoperative implantation path with the postoperative image coordinate system through the iterative nearest point algorithm. Using the planned path parameters as a reference, the screw implantation accuracy is classified into accuracy levels based on the deviation of the needle insertion point, angle deviation, depth deviation, and cortical penetration, and a postoperative report containing a three-dimensional comparison map of the path, evaluation conclusions, and improvement suggestions is generated.

Images