A deep brain stimulation path planning method

Abstract

The application discloses a deep brain stimulation path planning method, comprising the following steps: acquiring original medical data, and dividing training data set and test data set; data processing is performed on the divided data set; based on a data-driven model, a positioning result of a related structure is acquired, and key brain tissue positioning is completed through the positioning result; a symbol distance field calculation algorithm based on a hierarchical bounding box is adopted to determine a key position structure of the brain; a coordinate system of an electrode "safe implantation channel" is constructed; the constructed coordinate system is adopted to determine an optimal electrode implantation track, and deep brain stimulation path planning is completed; by introducing a positioning method based on deep learning, the application realizes accurate identification of key brain area structures in DBS surgery; by adopting nnDetection, segmentation of STN and red nucleus and accurate detection of brain area endangered tissues are simultaneously realized; the positioning accuracy of the method is improved, the time consumption is shortened, and the effect is improved.

Description

Technical Field

[0001] This invention belongs to the field of medical image processing and application technology, specifically relating to a deep brain stimulation pathway planning method. Background Technology

[0002] Deep brain stimulation (DBS) is a landmark new technology that uses tiny electrodes implanted in the brain to emit electrical pulses to stimulate specific nuclei, thereby obtaining abnormalities and data on brain neural activity and processing them accordingly. Deep brain stimulation has important significance in both laboratory research and clinical practice.

[0003] Deep brain stimulation requires the implantation of electrodes in the brain. These electrodes emit electrical pulses to stimulate specific nuclei. The electrode pathways are hidden within the complex brain structure, and implantation must effectively avoid cerebral blood vessels and critical brain structures. Therefore, the implantation pathways of the electrodes need to be planned carefully; otherwise, the implantation process carries significant risks.

[0004] Traditional DBS path planning is mainly done manually on multimodal images including fused CT and MRI. However, two-dimensional images cannot provide sufficient depth information. In this way, it is easy to make inaccurate target point judgment and find the most suitable puncture path. Surgical planning on three-dimensional graphics can provide more comprehensive information. However, relying on doctors to manually segment and find the optimal implantation path consumes a lot of time and energy.

[0005] In recent years, some algorithms have supported STN-DBS preoperative planning. These systems can provide better 2D or 3D visualization results, providing a promising direction for reducing doctors' workload and ensuring accurate surgical planning.

[0006] Achieving accurate STN-DBS surgical path planning involves two main steps: first, identifying an optimal stimulation target to alleviate the patient's symptoms; and second, finding an ideal implantation path that effectively avoids cerebral blood vessels and key brain structures, which must be done under multiple constraints.

[0007] In terms of locating key brain tissues, this anatomical information can help doctors accurately locate STN targets and identify the locations of tissues that must be avoided during puncture, such as sulci and ventricles. However, because the STN is small and located deep in the patient's brain, it is difficult to accurately locate the target manually by relying solely on preoperative MRI or CT images.

[0008] Currently, researchers have proposed various methods for locating STN targets, categorized into direct and indirect methods. Traditional indirect targeting methods utilize the midpoint of the line connecting the anterior commissure (AC) and posterior commissure (PC) to aid in STN target localization. In these methods, researchers often seek to automatically identify the AC and PC to improve aiming accuracy and also employ atlases-based registration models. In contrast, direct targeting involves using MRI technology to assist neurosurgeons in extracting morphological features for precise target localization; subsequently, image processing methods are used for direct visualization of the target region. However, these target localization methods are typically employed when STN boundaries are unclear, potentially leading to inaccurate target localization. Such inaccuracies are unacceptable in clinical practice because they increase the risk of intraoperative intracranial hemorrhage, resulting in poor postoperative treatment outcomes.

[0009] Therefore, clear target area boundaries are urgently needed to achieve precise implantation of STN-DBS surgical electrodes. Medical image segmentation technology allows doctors to better obtain morphological information of the STN target area. In clinical applications, the STN and red nucleus can assist doctors in locating the optimal target point within the STN. Recently, data-driven methods based on deep learning as the main approach for medical image segmentation have significantly reduced the workload of neurosurgeons. UNet-based models, employing an encoder-decoder architecture, have greatly improved the accuracy of computer vision and segmentation tasks. For microkernel segmentation, attention gates (AG) and recurrent residual U-Net are used to extract salient and low-dimensional features, respectively. Furthermore, Transformer provides a reliable method for extracting global features. Park et al. proposed a fully convolutional neural network for STN segmentation; Liu et al. proposed HA-UNet to segment the STN and Red nucleus, assisting in target localization. However, sulci and gyri are widely distributed across the entire brain surface, and the geometric features of the sulci are not significant, with variations in their distribution among individual patients. Therefore, the above methods face challenges in segmenting gyri and sulci. On the other hand, once the stimulation point is identified, the clinician surgeon must guide the electrode to its final location while avoiding puncturing dangerous structures such as sulci and ventricles during surgery. When reaching the solid surface, the electrode can only pass through the gyri; it is worth noting that during surgery, the surgeon only needs to determine the location of the sulci or ventricles and avoid puncturing them, which means that detecting the location of these structures is sufficient, without the need for detailed texture information; in this context, nnUNet and nnDetection have demonstrated their high performance in various scenarios, such as object detection and segmentation, showing potential for precise localization of STN targets and critical brain structures at risk.

[0010] In preoperative path planning for STN-DBS, another crucial stage is establishing the optimal electrode insertion trajectory from the skull entry point to the stimulation target, while avoiding puncture of critical structures. The planning results are essential for preventing intraoperative intracranial hemorrhage and reducing postoperative complications. Many trajectory planning methods have been developed for PD treatment: Essert et al. introduced a set of geometric constraints to draw the DBS planned insertion path; Dergachyova et al. employed a novel stimulation point to implement their algorithm; Bai et al. proposed a preoperative path planning pipeline and compared the planning results with clinicians' electrode implantation trajectories; Yao et al. used a random forest algorithm to correlate intraoperative microelectrode recordings (MER) with preoperative MRI, and then used the correlation between MER and MRI to deviate from the electrode trajectory. Nevertheless, clinicians typically rely on the midpoint of the AC-PC line to determine the electrode implantation path; the inconsistency between the established graphical coordinate system and the clinical planning coordinate system hinders the practical application of these methods.

[0011] On the other hand, flexible needles, as a more adaptable implantation method, have begun to attract attention in recent years. Segato et al. proposed an inverse reinforcement learning preoperative path planning method that seamlessly integrates the planning framework with user interaction. Their simulations on the human brain dataset demonstrated the clinical application potential of STN-DBS surgery. Zhao et al. proposed a comprehensive model that combines needle defect modeling, insertion angle correction, and trajectory optimization for preoperative planning in multi-layered tissues. Many researchers are also dedicated to soft tissue deformable needle implantation, striving to find suitable paths. These implantation methods allow for more suitable pathways within the empirical channel, providing a direction for the future development of STN-DBS surgical path planning. However, due to their non-rigid nature, these methods require a significant amount of time to calculate suitable implantation paths for preoperative path planning. Furthermore, surgeons need to guide the electrodes very carefully during the procedure, which severely prolongs the clinical surgery time.

[0012] In summary, most current methods for STN-DBS surgical path planning suffer from insufficient positioning accuracy and excessively long path planning time. Summary of the Invention

[0013] The purpose of this invention is to provide a deep brain stimulation path planning method that improves positioning accuracy and shortens path planning time.

[0014] The deep brain stimulation pathway planning method provided by this invention includes the following steps:

[0015] S1. Obtain the raw medical data and divide it into training and test datasets;

[0016] S2. Perform data processing on the dataset divided in step S1;

[0017] S3. Using the dataset processed in step S2, based on the data-driven model, obtain the localization results of relevant structures, and complete the localization of key brain tissues using the localization results;

[0018] S4. A symbolic distance field calculation algorithm based on hierarchical bounding boxes is used to determine key brain structures;

[0019] S5. Using the key location structure determined in step S4, construct the coordinate system of the electrode "safe implantation channel";

[0020] S6. Using the coordinate system constructed in step S5, determine the optimal electrode implantation trajectory and complete the planning of the deep brain stimulation path;

[0021] Step S1, which involves acquiring raw medical data and dividing it into training and testing datasets, specifically includes:

[0022] Obtain all diagnostic and treatment datasets;

[0023] The original medical data included preoperative 3T magnetic resonance imaging (MRI) scans (T1-weighted and T2-weighted), preoperative computed tomography (CT) scans, and postoperative retrospective CT images; meanwhile, the basal true values ​​of the gyri, sulci, ventricles, subthalamic nucleus (STN), and red nucleus were manually marked on each MRI slice.

[0024] According to the set number, samples are randomly selected from the acquired raw medical data as the training dataset, and the remaining samples are selected as the test dataset.

[0025] Step S2, which involves data processing of the dataset divided in step S1, specifically includes:

[0026] Five-fold cross-validation is performed on the training dataset, and the model and segmentation results are validated on the test dataset using five-fold cross-validation.

[0027] The performance of the method is evaluated by averaging the results on the test dataset;

[0028] Step S3 involves using the dataset processed in step S2, based on a data-driven model, to obtain the localization results of relevant structures, and then using these results to locate key brain tissues. Specifically, this includes:

[0029] In the preprocessing stage, key brain structures are manually delineated. The portion from the basal fissure to the boundary of the convex region is marked as a sulcus within the cortex, and the remaining portion is marked as a gyrus. Baseline ground truth labels are defined based on the contours of the ventricles, STN, and red nucleus. The locations of the sulci, gyrus, and ventricles are determined using detection branches. The segmentation results of the STN and red nucleus are obtained using the segmentation branches of the nnDetection model. The optimal stimulation point within the STN is semi-automatically located using the boundaries of the STN and red nucleus. The detection results of the sulci, gyrus, and ventricles are used as constraints for STN-DBS implantation path planning.

[0030] Step S4, which uses a symbolic distance field calculation algorithm based on hierarchical bounding boxes to determine key brain structures, specifically includes:

[0031] The following formula represents a given initial triangular input mesh:

[0032] in, Represents the signed distance function; Represents any triangular network; Represents the unsigned distance field; Representing a triangular network Approximate value; Represents a three-dimensional spatial domain. ;; The approximate function is shown in the following formula:

[0033] in, , indicating that the point on the grid is located internal;

[0034] pass The calculation formula is used to deduce the field of each point in the triangular mesh. The symbolic distance field (SDF) value is calculated and saved to the corresponding 3D mapping.

[0035] Integrate the positions of each structure according to the set proportions: Select the actual size of the patient's skull as the final fusion image, and generate the SDF values ​​of all other structures proportionally based on the selected skull proportions; merge all values ​​into the skull's SDF image; select... This represents the SDF file after the various parts have been merged. Indicates the SDF of the patient's skull; SDF representing brain sulci; SDF representing the gyri; SDF representing STN;

[0036] Step S5, which involves constructing a coordinate system for the electrode "safe implantation channel" using the key location structure determined in step S4, specifically includes:

[0037] A transformation is performed on the CT coordinate system to align it with the clinical coordinate system: given a fixed point in the CT coordinate system, the position of the corresponding point in the clinical planning coordinate system is found; the rotation matrix of the fixed point is calculated. and displacement matrix The final conversion result is obtained;

[0038] Select any point in the CT coordinate system, and calculate the position of the corresponding point in the clinical planning coordinate system using the same rotation and translation matrix. The calculation formula is shown below:

[0039] in, This indicates the position of any point in the clinical coordinate system; This indicates the corresponding position of the point in the CT coordinate system;

[0040] The positions of the obtained three-dimensional points in the CT and clinical planning coordinate systems are marked, and the midpoint of the line between the midpoint of the anterior commissure (AC) and the midpoint of the posterior commissure (PC) is selected.

[0041] Establish a coordinate system with the midpoint of the selected AC-PC line as the origin;

[0042] Step S6, which uses the coordinate system constructed in step S5 to determine the optimal electrode implantation trajectory and complete the deep brain stimulation path planning, specifically includes:

[0043] Transform the electrode implantation task into solving an energy function. The energy optimization problem is described by the following formula:

[0044] in, Indicates a hard constraint; Indicates soft constraint 1; Indicates soft constraint 2; Indicates soft constraint 3;

[0045] Using both hard and soft constraints, the electrodes are safely inserted from the insertion point on the skull. Pierce the target point of STN ;

[0046] The insertion path set is represented by the following formula:

[0047] in, This represents the set of insertion paths, which is the first... Target points To the Insertion points The path; , This represents the total number of potential paths within the "safe implantation channel" coordinate system;

[0048] The first step in the energy optimization problem of energy functions is to determine the specific anatomical structures that cannot be punctured, including cerebral blood vessels and ventricles;

[0049] If the hard constraint satisfies the condition that it neither punctures the sulci nor the ventricles, then: ;

[0050] If any of the sulci or ventricles in the brain are punctured, then: ;

[0051] 1) Calculate soft constraint 1:

[0052] The soft constraint 1 is calculated using the following formula. :

[0053] in, Indicates the distance between two points; Indicates the insertion path; Indicate the target point; express Discrete points on the path; Indicates the nearest point on the nearest obstacle; This represents the total number of discrete points;

[0054] pass The constraint insertion path maintains a distance from the critical structure within a set range;

[0055] 2) Calculate soft constraint 2:

[0056] Increase the contact area between the implanted electrode and the STN region by minimizing the angle between the planned entry point and the relevant plane normal direction;

[0057] The soft constraint 2 is calculated using the following formula. :

[0058] in, This indicates the angle between the planned trajectory and the direction of the skull normal; This represents the angle between the planned trajectory and the direction of the normal to the gyrus;

[0059] 3) Calculate soft constraint 3:

[0060] Establish limits on electrode implantation length within a defined range;

[0061] Calculate the closest distance between the skull and the target. ;

[0062] The soft constraint 3 is calculated using the following formula. :

[0063] in, This indicates the distance between the target point and the entry point into the skull; Indicates the minimum allowable length of the implanted electrode; Indicates the maximum allowable length of the implanted electrode;

[0064] An electrode implantation path optimization algorithm based on multiple medical constraints is used to determine the minimum energy function among all candidate entry points that satisfy all the above constraints. The corresponding entry point is determined, and the smallest entry point is taken as the optimal insertion point.

[0065] 4) Optimal electrode implantation trajectory planning process:

[0066] An electrode implantation path optimization algorithm based on multiple medical constraints is used to describe the optimal electrode implantation trajectory planning process, which specifically includes:

[0067] Input: 3D model of the skull 3D model of the brain gyri 3D model of obstacle avoidance organization Target location on STN ;

[0068] Output: Optimal electrode implantation direction , Location of the entry point on the patient's skull ;

[0069] (1) Initialization process:

[0070] Initialize the direction on the axial plane The distance from the target point to the needle entry point in the vertical direction of the patient's skull. ;

[0071] (2) Total energy calculation:

[0072] pass , , , Calculate the total energy ;

[0073] use calculate The calculation formula is as follows:

[0074] When the number of iterations Less than the maximum number of iterations At that time, utilize , , , Alternative in soft constraint formula , , The corresponding position is calculated using the following formula. :

[0075] when When, select the current one. Replace the optimal electrode implantation direction Choose the one obtained Replacement total energy ;

[0076] when At that time, keep Continue iterating until... No longer getting smaller, or the number of iterations Reaching the maximum number of iterations When the iteration ends;

[0077] (3) Update processing:

[0078] Update the optimal distance between the target point and the skull. Implantation direction .

[0079] This invention provides a deep brain stimulation (DBS) path planning method that achieves precise identification of key brain regions during DBS surgery by introducing a deep learning-based localization method. It employs a segmentation network structure to simultaneously segment the STN and red nucleus, as well as accurately detect brain regions at risk. An empirical channel coordinate system is constructed, enabling the planning results to be directly applied clinically, providing surgeons with a more effective and safer method for implantation surgery. By establishing a clinical experience coordinate system, the electrode implantation range agreed upon by experts is directly used as a constraint in the algorithm, reducing the computation time for the optimal implantation path and giving the algorithm superior time efficiency. The method of this invention improves localization accuracy, reduces time consumption, and enhances effectiveness. Attached Figure Description

[0080] Figure 1This is a schematic diagram of the method flow of the present invention.

[0081] Figure 2 This is a schematic diagram illustrating the implementation process of the method of the present invention.

[0082] Figure 3 This diagram illustrates the construction of the electrode "safe implantation channel" coordinate system and the conversion between the CT and clinical coordinate systems in the method of this invention.

[0083] Figure 4 This is a schematic diagram showing the comparison results of electrode implantation paths obtained by three different methods in the present invention under a clinical DBS surgical planning system.

[0084] Figure 5 This diagram illustrates the comparison results of three different methods selected in this invention applied to a clinical medication planning system. Detailed Implementation

[0085] like Figure 1 The diagram shown is a flowchart of the method of the present invention. Figure 2 The following is a schematic diagram of the implementation process of the method of the present invention: The deep brain stimulation path planning method provided by the present invention includes the following steps:

[0086] S1. Obtain the raw medical data and divide it into training and testing datasets; specifically including:

[0087] Obtain all diagnostic and treatment datasets;

[0088] The method of this invention selects to obtain all diagnostic and treatment datasets for Parkinson's disease patients;

[0089] The original medical data included preoperative 3T magnetic resonance imaging (MRI) scans (T1-weighted and T2-weighted), preoperative computed tomography (CT) scans, and postoperative retrospective CT images; meanwhile, the basal true values ​​of the gyri, sulci, ventricles, subthalamic nucleus (STN), and red nucleus were manually marked on each MRI slice.

[0090] According to the set number, samples are randomly selected from the acquired raw medical data as the training dataset, and the remaining samples are selected as the test dataset.

[0091] In the method of this invention, a total of 100 Parkinson's disease patients' diagnostic and treatment datasets were collected. The patients' ages ranged from 23 to 76 years, and the average disease duration was 7.18 years. At the same time, a neurosurgeon with more than 5 years of experience was hired to manually mark the basal true values ​​of the gyri, sulci, ventricles, STN and red nucleus on each MRI slice.

[0092] Eighty samples were randomly selected to form the training dataset, and the remaining samples formed the test dataset.

[0093] S2. Perform data processing on the dataset divided in step S1; specifically including:

[0094] Five-fold cross-validation is performed on the training dataset, and the model and segmentation results are validated on the test dataset using five-fold cross-validation.

[0095] The performance of the method is evaluated by averaging the results on the test set;

[0096] In the method of this invention, 8 patients with treatment results were selected. Each patient underwent the complete surgical procedure and had a good prognosis. All patients had bilateral DBS electrodes implanted. The examination period was from March 2023 to June 2023. All patients met the criteria for idiopathic Parkinson's disease from the British Parkinson's Disease Association Brain Bank. The male-to-female ratio was 1:1, the average age was 64.25 years, and the average disease duration was 8.0 years.

[0097] S3. Using the dataset processed in step S2, based on a data-driven model, obtain the localization results of relevant structures, and complete the localization of key brain tissues using the localization results; specifically including:

[0098] By maximizing the contact area between the inserted electrode and the STN target area, patients can achieve the most favorable treatment results.

[0099] In the preprocessing stage, key brain structures are manually delineated. The portion from the basal fissure to the boundary of the convex region is marked as a sulcus within the cortex, and the remaining portion is marked as a gyrus. Baseline ground truth labels are defined based on the contours of the ventricles, STN, and red nucleus. The locations of the sulci, gyrus, and ventricles are determined using detection branches. The segmentation results of the STN and red nucleus are obtained using the segmentation branches of the nnDetection model. The optimal stimulation point within the STN is semi-automatically located using the boundaries of the STN and red nucleus. The detection results of the sulci, gyrus, and ventricles are used as constraints for STN-DBS implantation path planning.

[0100] S4. A symbolic distance field calculation algorithm based on hierarchical bounding boxes is used to determine key brain structures; specifically including:

[0101] The following formula represents a given initial triangular input mesh:

[0102] in, Represents the signed distance function; Represents any triangular network; Represents the unsigned distance field; Representing a triangular network Approximate value; Represents a three-dimensional spatial domain. ;; The approximate function is shown in the following formula:

[0103] in, , indicating that the point on the grid is located internal;

[0104] pass The calculation formula is used to deduce the field of each point in the triangular mesh. The symbolic distance field (SDF) value is calculated and saved to the corresponding 3D mapping.

[0105] Integrate the positions of each structure according to the set proportions: Select the actual size of the patient's skull as the final fusion image, and generate the SDF values ​​of all other structures proportionally based on the selected skull proportions; merge all values ​​into the skull's SDF image; select... This represents the SDF file after the various parts have been merged. Indicates the SDF of the patient's skull; SDF representing brain sulci; SDF representing the gyri; SDF representing STN;

[0106] S5. Using the key location structure determined in step S4, construct the coordinate system for the electrode "safe implantation channel"; specifically including:

[0107] like Figure 3 The diagram shows the construction of the electrode "safe implantation channel" coordinate system and the conversion between the CT and clinical coordinate systems in the method of this invention:

[0108] A transformation is performed on the CT coordinate system to align it with the clinical coordinate system: given a fixed point in the CT coordinate system, the position of the corresponding point in the clinical planning coordinate system is found; the rotation matrix of the fixed point is calculated. and displacement matrix The final conversion result is obtained;

[0109] Select any point in the CT coordinate system, and calculate the position of the corresponding point in the clinical planning coordinate system using the same rotation and translation matrix. The calculation formula is shown below:

[0110] in, This indicates the position of any point in the clinical coordinate system; This indicates the corresponding position of the point in the CT coordinate system;

[0111] The positions of the obtained three-dimensional points in the CT and clinical planning coordinate systems are marked, and the midpoint of the line between the midpoint of the anterior commissure (AC) and the midpoint of the posterior commissure (PC) is selected.

[0112] Establish a coordinate system with the midpoint of the selected AC-PC line as the origin;

[0113] S6. Using the coordinate system established in step S5, determine the optimal electrode implantation trajectory and complete the planning of the deep brain stimulation path; specifically including:

[0114] Transform the electrode implantation task into solving an energy function. The energy optimization problem is described by the following formula:

[0115] in, Indicates a hard constraint; Indicates soft constraint 1; Indicates soft constraint 2; Indicates soft constraint 3;

[0116] Using both hard and soft constraints, the electrodes are safely inserted from the insertion point on the skull. Pierce the target point of STN ;

[0117] The insertion path set is represented by the following formula:

[0118] in, This represents the set of insertion paths, which is the first... Target points To the Insertion points The path; , This represents the total number of potential paths within the "safe implantation channel" coordinate system;

[0119] The first step in the energy optimization problem of energy functions is to determine the specific anatomical structures that cannot be punctured, including cerebral blood vessels and ventricles;

[0120] If the hard constraint satisfies the condition that it neither punctures the sulci nor the ventricles, then: ;

[0121] If any of the sulci or ventricles in the brain are punctured, then: ;

[0122] 1) Calculate soft constraint 1:

[0123] The soft constraint 1 is calculated using the following formula. :

[0124] in, Indicates the distance between two points; Indicates the insertion path; Indicate the target point; express Discrete points on the path; Indicates the nearest point on the nearest obstacle; This represents the total number of discrete points;

[0125] pass The constraint insertion path maintains a distance from the critical structure within a set range;

[0126] 2) Calculate soft constraint 2:

[0127] Increase the contact area between the implanted electrode and the STN region by minimizing the angle between the planned entry point and the relevant plane normal direction;

[0128] The soft constraint 2 is calculated using the following formula. :

[0129] in, This indicates the angle between the planned trajectory and the direction of the skull normal; This represents the angle between the planned trajectory and the direction of the normal to the gyrus;

[0130] 3) Calculate soft constraint 3:

[0131] Establish limits on electrode implantation length within a defined range;

[0132] Calculate the closest distance between the skull and the target. In the method of this invention, the following is selected: ;

[0133] The soft constraint 3 is calculated using the following formula. :

[0134] in, This indicates the distance between the target point and the entry point into the skull; Indicates the minimum allowable length of the implanted electrode; Indicates the maximum allowable length of the implanted electrode;

[0135] An electrode implantation path optimization algorithm based on multiple medical constraints is used to determine the minimum energy function among all candidate entry points that satisfy all the above constraints. The corresponding entry point is determined, and the smallest entry point is taken as the optimal insertion point.

[0136] 4) Optimal electrode implantation trajectory planning process:

[0137] An electrode implantation path optimization algorithm based on multiple medical constraints is used to describe the optimal electrode implantation trajectory planning process, which specifically includes:

[0138] Input: 3D model of the skull 3D model of the brain gyri 3D model of obstacle avoidance organization Target location on STN ;

[0139] Output: Optimal electrode implantation direction , Location of the entry point on the patient's skull ;

[0140] (1) Initialization process:

[0141] Initialize the direction on the axial plane The distance from the target point to the needle entry point in the vertical direction of the patient's skull. ;

[0142] (2) Total energy calculation:

[0143] pass , , , Calculate the total energy ;

[0144] use calculate The calculation formula is as follows:

[0145] When the number of iterations Less than the maximum number of iterations At that time, utilize , , , Alternative in soft constraint formula , , The corresponding position is calculated using the following formula. :

[0146] when When, select the current one. Replace the optimal electrode implantation direction Choose the one obtained Replacement total energy ;

[0147] when At that time, keep Continue iterating until... No longer getting smaller, or the number of iterations Reaching the maximum number of iterations When the iteration ends;

[0148] (3) Update processing:

[0149] Update the optimal distance between the target point and the skull. Implantation direction ;

[0150] In this invention, the method was experimentally verified using retrospective data from 8 clinical patients. Simultaneously, a clinician was invited to simulate electrode implantation using both the system's built-in planning algorithm and the algorithm of this invention within a commonly used DBS surgical planning system. The distances between the electrodes and the sulci and ventricles, as well as the surgical implantation results, were compared between the two methods. Furthermore, the feasibility of this invention was evaluated by the clinician.

[0151] For each patient's data, the surgeon first performs a preprocessing step, using preoperative MRI data (T1w) as a reference modality. The preoperative and postoperative CT data are registered and fused with the preoperative MRI on the system. Key structures within the skull and brain are simultaneously observed on the obtained fused images. The brain structure is semi-automatically reconstructed using a navigation system and applied to STN-DBS surgery. The surgeon visually locates and marks the AC and PC positions, then determines the midline between the two hemispheres, completing the construction of the planning coordinate system. Subsequently, the system can generate planning rules for bidirectional electrodes. The electrode planning results obtained through this invention are imported into the system. In the flight system, to verify the performance of the three methods, the construction of the empirical channel coordinate system enabled the electrode planning results obtained by the method of this invention to be directly compared with the other two results in the planning system. In the final experimental verification, preoperative CT was used to display the visualization results of the system's built-in algorithm and the method of this invention; postoperative CT was used to display the actual implantation results; MRI was used to measure the closest distances between the bidirectional electrode planning results obtained by the above three different methods and the sulci and ventricles, respectively; the doctor measured the closest sulci and ventricles distances of the two needle paths of the three methods by combining sagittal, coronal and axial views on the preoperative and postoperative fused images in a layer-by-layer manner.

[0152] like Figure 4 The diagram shows a comparison of electrode implantation paths obtained through three different methods in the present invention within a clinical DBS surgical planning system. Figure 5 The diagram shows a comparison of three different methods selected in this invention applied to a clinical medication planning system: In clinical practice, the electrode diameter is 1.3 mm. Based on clinicians' recommendations, this invention sets the safe implantation radius to 1.20 mm, leaving a 0.55 mm adjustment range excluding the electrode radius. Figure 5 It can be seen that many planning results generated by the built-in algorithm of the planning system do not meet clinical requirements. The main reason for this is that the built-in algorithm generates planning results on a pre-set template. For each case, the electrode implantation angle is consistent by default. However, since the brain structure of each patient is not completely consistent, the distance between the electrode generated in this way and the patient's sulci and ventricles is less than the safe threshold. As a result, doctors need to make adjustments based on the planning results generated by the above algorithm before they can be used in clinical practice.

[0153] Figure 5 The three-dimensional visualization results shown further support the claims of the method of the present invention; Figure 5 Figures A-C in the table show visualizations from three different perspectives obtained from various methods applied to implantation trajectories in a widely used clinical planning system. Electrode 1 corresponds to the result generated by the system's native algorithm, electrode 2 represents the actual implantation result, and electrode 3 represents the result obtained using the method of this invention. Figures D-G show the measurement results of the distances between the left and right electrodes and the sulci and ventricles determined by the algorithm introduced within the system. Figures H-K show the measurement results of the distances between the actually implanted left and right electrodes and the sulci and ventricles, respectively. Figures L-O show the distances between the left and right electrode trajectories and the sulci and ventricles determined by the method of this invention.

[0154] When the results of the planning system's built-in algorithm are compared with the clinical implantation results and the results of the method of this invention, the system's planning results show that the obtained results are too close to the coronal suture line, resulting in the results not being accepted by clinicians. In addition, the method of this invention also calculates the average distance between the electrode and the obstacle obtained by the method of this invention: the average distance between the right electrode and the obstacle is 2.36 mm (sulcus) and 6.64 mm (ventricle), respectively, and the average distance between the left electrode and the obstacle is 1.76 mm and 5.22 mm, respectively. Compared with the clinical planning method that relies on surgeons to determine the electrode implantation trajectory through repeated trials, this invention provides a more effective and safer method for STN-DBS electrode insertion path planning.

Claims

1. A deep brain stimulation pathway planning method, comprising the following steps: S1. Obtain the raw medical data and divide it into training and test datasets; S2. Perform data processing on the dataset divided in step S1; S3. Using the dataset processed in step S2, based on a data-driven model, obtain the localization results of relevant structures, and complete the localization of key brain tissues using the localization results; specifically including: In the preprocessing stage, key brain structures are manually delineated. The portion from the basal fissure to the boundary of the convex region is marked as a sulcus within the cortex, and the remaining portion is marked as a gyrus. Baseline ground truth labels are defined based on the contours of the ventricles, STN, and red nucleus. The locations of the sulci, gyrus, and ventricles are determined using detection branches. The segmentation results of the STN and red nucleus are obtained using the segmentation branches of the nnDetection model. The optimal stimulation point within the STN is semi-automatically located using the boundaries of the STN and red nucleus. The detection results of the sulci, gyrus, and ventricles are used as constraints for STN-DBS implantation path planning. S4. A symbolic distance field calculation algorithm based on hierarchical bounding boxes is used to determine key brain structures; S5. Using the key location structure determined in step S4, construct the coordinate system for the electrode "safe implantation channel"; specifically including: A transformation is performed on the CT coordinate system to align it with the clinical coordinate system: given a fixed point in the CT coordinate system, the position of the corresponding point in the clinical planning coordinate system is found; the rotation matrix of the fixed point is calculated. and displacement matrix The final conversion result is obtained; Select any point in the CT coordinate system, and calculate the position of the corresponding point in the clinical planning coordinate system using the same rotation and translation matrix. The calculation formula is shown below: in, This indicates the position of any point in the clinical coordinate system; This indicates the corresponding position of the point in the CT coordinate system; The positions of the obtained three-dimensional points in the CT and clinical planning coordinate systems are marked, and the midpoint of the line between the midpoint of the anterior commissure (AC) and the midpoint of the posterior commissure (PC) is selected. Establish a coordinate system with the midpoint of the selected AC-PC line as the origin; S6. Using the coordinate system constructed in step S5, determine the optimal electrode implantation trajectory and complete the planning of the deep brain stimulation path.

2. The deep brain stimulation path planning method according to claim 1, characterized in that... Step S1, which involves acquiring raw medical data and dividing it into training and testing datasets, specifically includes: Obtain all diagnostic and treatment datasets; The original medical data included preoperative 3T MRI scans, preoperative computed tomography scans, and postoperative retrospective CT images; the preoperative 3T MRI scans included t1-weighted and t2-weighted images; at the same time, the basal true values ​​of the gyri, sulci, ventricles, subthalamic nucleus STN, and red nucleus were manually marked on each MRI slice. According to the set number, samples are randomly selected from the acquired raw medical data to serve as the training dataset, and the remaining samples are selected as the test dataset.

3. The deep brain stimulation path planning method according to claim 2, characterized in that... Step S2, which involves data processing of the dataset divided in step S1, specifically includes: Five-fold cross-validation is performed on the training dataset, and the model and segmentation results are validated on the test dataset using five-fold cross-validation. The performance of the method is evaluated by averaging the results on the test set.

4. The deep brain stimulation path planning method according to claim 3, characterized in that... Step S4, which uses a symbolic distance field calculation algorithm based on hierarchical bounding boxes to determine key brain structures, specifically includes: The following formula represents a given initial triangular input mesh: in, Represents the signed distance function; Represents any triangular network; Represents the unsigned distance field; Representing a triangular network Approximate value; Represents a three-dimensional spatial domain. ; The approximate function is shown in the following formula: in, , indicating that the point on the grid is located internal; pass The calculation formula is used to deduce the field of each point in the triangular mesh. The symbolic distance field (SDF) value is calculated and saved to the corresponding 3D mapping. Integrate the positions of each structure according to the set proportions: Select the actual size of the patient's skull as the final fusion image, and generate the SDF values ​​of all other structures proportionally based on the selected skull proportions; merge all values ​​into the skull's SDF image; select... This represents the SDF file after the various parts have been merged. Indicates the SDF of the patient's skull; SDF representing brain sulci; SDF representing the gyri; SDF representing STN.

5. The deep brain stimulation path planning method according to claim 4, characterized in that... Step S6, which uses the coordinate system constructed in step S5 to determine the optimal electrode implantation trajectory and complete the deep brain stimulation path planning, specifically includes: Transform the electrode implantation task into solving an energy function. The energy optimization problem is described by the following formula: in, Indicates a hard constraint; Indicates soft constraint 1; Indicates soft constraint 2; Indicates soft constraint 3; Using both hard and soft constraints, the electrodes are safely inserted from the insertion point on the skull. Pierce the target point of STN ; The insertion path set is represented by the following formula: in, This represents the set of insertion paths, which is the first... Target points To the Insertion points The path; , This represents the total number of potential paths within the "safe implantation channel" coordinate system; The first step in the energy optimization problem of energy functions is to determine the specific anatomical structures that cannot be punctured, including cerebral blood vessels and ventricles; If the hard constraint satisfies the condition that it neither punctures the sulci nor the ventricles, then: ; If any of the sulci or ventricles in the brain are punctured, then: ; 1) Calculate soft constraint 1: The soft constraint 1 is calculated using the following formula. : in, Indicates the distance between two points; Indicates the insertion path; Indicates the target point; express Discrete points on the path; Indicates the nearest point on the nearest obstacle; This represents the total number of discrete points; pass The constraint insertion path maintains a distance from the critical structure within a set range; 2) Calculate soft constraint 2: Increase the contact area between the implanted electrode and the STN region by minimizing the angle between the planned entry point and the relevant plane normal direction; The soft constraint 2 is calculated using the following formula. : in, This indicates the angle between the planned trajectory and the direction of the skull normal; This represents the angle between the planned trajectory and the direction of the normal to the gyrus; 3) Calculate soft constraint 3: Establish limits on electrode implantation length within a defined range; Calculate the closest distance between the skull and the target. ; The soft constraint 3 is calculated using the following formula. : in, This indicates the distance between the target point and the entry point into the skull; Indicates the minimum allowable length of the implanted electrode; Indicates the maximum allowable length of the implanted electrode; An electrode implantation path optimization algorithm based on multiple medical constraints is used to determine the energy function from all candidate entry points that satisfy all the above constraints. The minimum corresponding entry point is determined, and the determined minimum entry point is used as the optimal insertion point.

6. The deep brain stimulation path planning method according to claim 5, characterized in that... The optimal electrode implantation trajectory planning process specifically includes: An electrode implantation path optimization algorithm based on multiple medical constraints is used to describe the optimal electrode implantation trajectory planning process, which specifically includes: Input: 3D model of the skull 3D model of the brain gyri 3D model of obstacle avoidance organization Target location on STN ; Output: Optimal electrode implantation direction , Location of the entry point on the patient's skull ; (1) Initialization process: Initialize the direction on the axial plane The distance from the target point to the needle entry point in the vertical direction of the patient's skull. ; (2) Total energy calculation: pass , , , Calculate the total energy ; use calculate The calculation formula is as follows: When the number of iterations Less than the maximum number of iterations At that time, utilize , , , Alternative in soft constraint formula , , The corresponding position is calculated using the following formula. : when When, select the current one. Replace the optimal electrode implantation direction Choose the one obtained Replacement total energy ; when At that time, keep Continue iterating until... No longer getting smaller, or the number of iterations Reaching the maximum number of iterations When the iteration ends; (3) Update processing: Update the optimal distance between the target point and the skull. Implantation direction .

Images