A dynamic neural reconstruction tracking method based on the principle of obstacle avoidance

Abstract

A dynamic neural reconstruction tracking method based on the obstacle avoidance principle. First, an obstacle model is constructed according to the shape and size of the tumor. Secondly, the interference matrix of the fiber at different positions is calculated through the obstacle model. Then, the interference matrix is used to change the direction of fiber tracking so that the final fiber can avoid the tumor. Among them, the solution of the interference matrix is constrained by the fiber direction distribution and clinical anatomical structure conditions. Finally, the optimal parameters of the interference matrix are calculated to minimize the difference between the perturbed fiber direction and the estimated direction. In addition, the tumor shape and size used to define the interference matrix are obtained by modeling the obstacle through fitting the surface and optimizing the parameters using the polynomial approximation algorithm. The present invention can accurately reconstruct while bypassing the tumor during the fiber reconstruction process, and finally make the fiber tracking result effective and provide a surgical reference for neurosurgeons.

Description

Technical Field

[0001] The present invention belongs to the fields of medical image processing and artificial intelligence, and is a dynamic nerve reconstruction tracking method based on the obstacle avoidance principle. Background Art

[0002] The nerve fiber reconstruction technology is extremely important for clinical neurosurgical operations. It helps to guide and assist neurosurgeons in removing brain tumors without damaging nerves. This role greatly improves the quality of life of patients after surgery on the one hand, and on the other hand, patients do not need to worry about the problem of functional defects caused by accidental injury to other nerves during the operation. The above two reasons make the nerve fiber reconstruction technology of extremely high value. In clinical patient data, tumor patients account for a relatively high proportion. Importantly, due to individual differences in each patient; different natures and different positions of tumors will have different impacts on nerves; the surgical approach strongly depends on the clinical experience of doctors. All the above reasons indicate that the reconstruction of peritumoral nerve fibers is necessary. During the operation, it can assist doctors in removing tumors without damaging other nerves, thereby protecting other functions of patients from being damaged. During the nerve fiber reconstruction process, due to the compression of the tumor on the surrounding nerves, the peritumoral nerves will be displaced or even greatly bent. This increases the difficulty of nerve fiber reconstruction. Especially the peritumoral edema area will cause errors in the estimation results of fiber direction distribution, thus affecting subsequent nerve fiber tracking. In addition, most current fiber tracking methods (such as probabilistic tracking and deterministic tracking) will cause fiber tracking errors or termination when tracking to the peritumoral area due to errors in fiber direction distribution estimation and premature satisfaction of termination conditions. Such a situation cannot provide effective and accurate information for clinical doctors. Therefore, developing an accurate, less affected by tumors, and robust nerve fiber reconstruction algorithm is still a necessary and challenging task. Summary of the Invention

[0003] In order to solve the problem that the existing nerve fiber tracking algorithm is greatly affected by tumors, and there are problems such as wrong tracking and premature termination in the nerve fiber tracking around the tumor, the present invention proposes a nerve fiber reconstruction method based on the obstacle avoidance principle, which can accurately reconstruct while bypassing the tumor during the reconstruction process, and finally make the fiber tracking result effective and provide a surgical reference for neurosurgeons.

[0004] The technical solution adopted by the present invention to solve its technical problems is:

[0005] A dynamic nerve reconstruction tracking method based on the obstacle avoidance principle, the method includes the following steps:

[0006] Step 1: After obtaining the dMRI data of clinical oncology patients, convert it into DWI data, preprocess the DWI data, perform eddy current correction on the DWI, then perform motion correction, and then combine the b0 image, T1 image and T2 image for alignment for EPI correction;

[0007] Step 2: Medical experts reconstruct the tumor on the patient's medical images and label the starting area and ending area, perform surface fitting through the polynomial approximation algorithm to obtain the tumor model, and based on the above model, calculate the interference matrix for each position of the fiber; select the direction of the voxel peak image in the starting area as the initial direction;

[0008] Step 3: Based on the interference matrix, combine the position information of fiber tracking, the current position and the fiber direction distribution information in the neighborhood to optimize each parameter in the interference matrix, aiming to make the tracking direction after interference conform to clinical anatomical knowledge;

[0009] Step 4: Calculate the fiber tracking direction at each step to completely reconstruct the nerve fiber.

[0010] Furthermore, in the above Step 2, the interference matrix satisfies the following conditions:

[0011] 1.1) When performing fiber tracking, if the current tracking point is not within the tumor influence range, then the nerve fiber tracking will be completely based on the information of the nerve fiber direction distribution, that is, at this time the interference matrix will be an identity matrix, which will not affect the tracking direction;

[0012] 1.2) When the current tracking point is close to the tumor range, the interference matrix generated by the tumor equation will correct the current tracking direction and the correction intensity increases as the distance from the tumor boundary decreases. The interference matrix equation is as follows P(ξ):

[0013]

[0014] Among them, I represents the identity matrix, n(ξ) represents the normal vector, t(ξ) represents the tangent vector, and ρ(ξ), σ(ξ) respectively represent the weights of the current velocity direction affected by n(ξ), y(ξ). The expressions of ρ(ξ), σ(ξ) are as follows

[0015]

[0016] ρ in formula (2) 0 , σ 0 In the obstacle avoidance principle, they are called the repulsive force coefficient and the tangential coefficient. In the process of fiber reconstruction, these two parameters actually control the influence of the tumor morphological characteristics on the selection of the fiber direction; the satisfaction function τ controls the tangential influence within the set threshold, and the satisfaction function is expressed as:

[0017]

[0018] Among them, g is a given positive constant, is a positive threshold, and v(ξ) is the direction of the current voxel peak image;

[0019] When approaching the tumor, the original direction is set to the direction from the current position to the termination region. This setting ensures that it can be tracked to the termination region. The finally perturbed fiber direction is expressed as:

[0020] v 2 (ξ) = P(ξ)v(ξ) (4)

[0021] According to this direction, a fixed step size is selected for the next step of tracking, and the direction of each step is calculated in turn until the termination region is reached.

[0022] Furthermore, in the third step, the process of dynamically optimizing the interference matrix is as follows:

[0023] In order to make the direction selection of each step closer to the fiber direction distribution and conform to anatomical knowledge, and to realize the automatic dynamic optimization of each variable in the interference matrix p generated by the tumor. First, a clustering operation is performed on the directions of the voxel peak images in the 26-neighborhood around the current position, and then the results are sorted in descending order; then, the directions corresponding to the voxels closest to the current position in the top 3 types of directions are taken, and this direction is regarded as the target direction to be approximated; secondly, by optimizing the repulsive force parameter, tangential direction weight ρ 0 , σ 0 in the interference matrix, and the function coefficient g, the estimated direction V new after being corrected by the interference matrix is made to have the smallest difference from the target direction V dir , that is:

[0024]

[0025] D(ξ) represents the distance between the current position and the center of the obstacle, M(ξ) represents the distance from the center of the obstacle to the surface at the current position, and thr represents the distance threshold set according to the shape and size of the tumor; it can ensure that there are no false positives in the nerve fiber reconstruction result within the tumor, and it follows the fiber direction distribution law and conforms to anatomical knowledge.

[0026] Preferably, when the nerve fiber tracking is far away from the tumor, it is not corrected by the obstacle equation; when approaching the tumor, the tumor is regarded as an obstacle and modeled. The tumor is represented by an equation. Let F(ξ) be:

[0027]

[0028] Among them, (x 0, y 0 , z 0 ) represents the tumor center coordinates, where a, b, c, d, e, and f are coefficients respectively. Let F(ξ) = C, and a surface fitting is performed through a polynomial approximation algorithm to construct an obstacle model.

[0029] In the present invention, an obstacle model is constructed according to the shape and size of the tumor. The interference matrix at different positions is calculated through the obstacle model to change the fiber tracking direction, so that the final fibers can avoid the tumor. Among them, the solution of the interference matrix is subject to the conditional constraints of the fiber direction distribution and the clinical anatomical structure; finally, the optimal parameters of the interference matrix are calculated to minimize the difference between the perturbed fiber direction and the estimated direction; in addition, the tumor shape and size used to define the interference matrix are obtained by fitting the surface and optimizing the parameters using a polynomial approximation algorithm to model the obstacle.

[0030] The beneficial effects of the present invention are mainly manifested in that it can accurately reconstruct while bypassing the tumor during the reconstruction process, and finally make the fiber tracking results effective and provide surgical reference for neurosurgeons. Description of the Drawings

[0031] Figure 1 is a flowchart of a dynamic nerve reconstruction tracking method based on the obstacle avoidance principle. Detailed Embodiments

[0032] The present invention will be further described below with reference to the drawings.

[0033] Refer to Figure 1 , a dynamic nerve reconstruction tracking method based on the obstacle avoidance principle, includes the following steps:

[0034] Step 1: When the dMRI data of a clinical tumor patient is obtained, it is converted into DWI data. Since the patient will inevitably move or be affected by factors such as eddy current effects during scanning, artifacts will be generated in the scanned images, and these artifacts will affect our subsequent tracking effects. Therefore, it is necessary to preprocess the DWI data. During the preprocessing process, first, eddy current correction is performed on the DWI, then motion correction is performed, and then combined with the b0 image, T1 image, and T2 image alignment for EPI correction to minimize the influence of artifacts as much as possible;

[0035] Step 2: Medical experts reconstruct the tumor on the patient's medical image and mark the starting area and the ending area. The purpose is to perform surface fitting through a polynomial approximation algorithm to obtain a tumor model. Based on the above model, the interference matrix at each position of the fiber is calculated. In addition, the direction of the voxel peak image in the starting area is selected as the initial direction;

[0036] In the above Step 2, the definition of the interference matrix

[0037] The interference matrix satisfies the following conditions:

[0038] 1.1) When performing fiber tracking, if the current tracking point is not within the tumor influence range, then the nerve fiber tracking will be completely based on the information of the nerve fiber direction distribution. That is, at this time, the interference matrix will be an identity matrix, which will not affect the tracking direction;

[0039] 1.2) When the current tracking point is close to the tumor range, the interference matrix generated by the tumor equation will correct the current tracking direction, and the intensity of the correction will increase as the distance from the tumor boundary decreases. The interference matrix equation is as follows P(ξ):

[0040]

[0041] Where, I represents the identity matrix, n(ξ) represents the normal vector, t(ξ) represents the tangent vector, and ρ(ξ), σ(ξ) represent the weights of the current velocity direction affected by n(ξ), t(ξ) respectively. The expressions of ρ(ξ), σ(ξ) are as follows

[0042]

[0043] ρ in formula (2) 0 , σ 0 In the obstacle avoidance principle, they are called the repulsive force coefficient and the tangential coefficient. In the fiber reconstruction process, these two parameters actually control the influence of the tumor morphological characteristics on the selection of the fiber direction. The function τ that satisfies is to control the tangential influence within the set threshold. The satisfaction function is expressed as:

[0044]

[0045] Where, g is a given positive constant, is the positive threshold, and v(ξ) is the direction of the current voxel peak image;

[0046] When approaching the tumor, the original direction is set to the direction from the current position to the termination area. This setting ensures that it can be tracked to the termination area. The finally perturbed fiber direction is expressed as:

[0047] v 2 (ξ) = P(ξ)v(ξ) (10)

[0048] According to this direction, a fixed step size is selected for the next step of tracking, and the direction of each step is calculated in turn until the termination area is reached;

[0049] Step 3: Based on the interference matrix, optimize each parameter in the interference matrix by combining the position information of fiber tracking, the current position, and the fiber direction distribution information within the neighborhood, with the aim of making the tracking direction after being interfered with conform to clinical anatomical knowledge;

[0050] In the said Step 3, the process of dynamically optimizing the interference matrix is as follows:

[0051] In order to make the direction selection in each step closer to the fiber direction distribution and conform to anatomical knowledge, and to realize the automatic dynamic optimization of each variable in the interference matrix p generated by the tumor. First, perform a clustering operation on the directions of the voxel peak images in the 26-neighborhood around the current position, and then sort the results in descending order; then, take the directions corresponding to the voxels closest to the current position in the top 3 types of directions, and regard this direction as the target direction to be approximated; secondly, by optimizing the repulsive force parameter, the tangential direction weight ρ 0 , σ 0 , and the function coefficient g that are satisfied, make the estimated direction V new after being corrected by the interference matrix have the smallest difference from the target direction V dir , that is:

[0052]

[0053] D(ξ) represents the distance between the current position and the center of the obstacle, M(ξ) represents the distance from the center of the obstacle to the surface under the premise of the current position, and thr represents the distance threshold set according to the tumor shape and size; it can ensure that there are no false positives in the nerve fiber reconstruction result within the tumor, and it follows the fiber direction distribution law and conforms to anatomical knowledge;

[0054] When the nerve fiber tracking is far away from the tumor, it is not corrected by the obstacle equation; when approaching the tumor, the tumor is regarded as an obstacle and modeled. The tumor is represented by an equation. Let F(ξ) be:

[0055]

[0056] Among them, (x 0 , y 0 , z 0 ) represents the coordinates of the tumor center, and a, b, c, d, e, f are coefficients respectively. Let F(ξ) = C, and construct the obstacle model by fitting the surface through the polynomial approximation algorithm.

[0057] Step 4: Calculate the fiber tracking direction for each step, so as to completely reconstruct the nerve fibers.

[0058] In this embodiment, first, when the dMRI data of a clinical oncology patient is obtained, it is converted into DWI data. Since the patient will inevitably move or be affected by factors such as eddy current effects during the scan, artifacts will be generated in the scanned images. These artifacts will affect our subsequent tracking effect. Therefore, it is necessary to preprocess the DWI data. During the preprocessing process, first, eddy current correction is performed on the DWI, then motion correction is carried out, and then, in combination with the b0 image, T1 image, and T2 image alignment, EPI correction is performed to minimize the influence of artifacts;

[0059] Secondly, medical experts reconstruct the tumor on the patient's medical images and mark the starting area and ending area. The purpose is to perform surface fitting through the polynomial approximation algorithm to obtain the tumor model. Based on the above model, the interference matrix at each position of the fiber is calculated. In addition, the direction of the voxel peak image in the starting area is selected as the initial direction;

[0060] Finally, based on the interference matrix obtained in the above steps, each parameter in the interference matrix is optimized by combining the position information of fiber tracking, the current position, and the fiber direction distribution information in the neighborhood, with the aim of making the tracking direction after being interfered with conform to clinical anatomical knowledge;

[0061] When the above three inputs, namely the tumor mask, the starting area mask, and the ending area mask, are prepared, the fiber tracking direction of each step is calculated through the above method, thereby completely reconstructing the nerve fiber.

Claims

1. A dynamic neural reconstruction tracking method based on the obstacle avoidance principle, characterized in that, the method comprises the following steps: Step 1: After obtaining the dMRI data of a clinical tumor patient, convert it into DWI data, preprocess the DWI data, perform eddy current correction on the DWI, then perform motion correction, and then perform EPI correction by aligning with the b0 image, T1 image and T2 image; Step 2: Medical experts reconstruct the tumor on the patient's medical image and mark the starting area and the ending area, perform surface fitting through the polynomial approximation algorithm to obtain the tumor model, and calculate the interference matrix at each position of the fiber based on the above model; select the direction of the voxel peak image in the starting area as the initial direction; Step 3: Based on the interference matrix, combine the position information of fiber tracking, the current position and the fiber direction distribution information in the neighborhood to optimize each parameter in the interference matrix, aiming to make the tracking direction after being interfered conform to clinical anatomical knowledge; Step 4: Calculate the fiber tracking direction at each step to completely reconstruct the nerve fiber; In the second step, the interference matrix satisfies the following conditions: 1.1) When performing fiber tracking, if the current tracking point is not within the tumor influence range, then the nerve fiber tracking will completely depend on the nerve fiber direction distribution information, that is, at this time the interference matrix will be an identity matrix, which will not affect the tracking direction; 1.2) When the current tracking point is close to the tumor range, the interference matrix generated by the tumor equation will correct the current tracking direction and the correction intensity increases as the distance from the tumor boundary decreases. The interference matrix equation is as follows P(ξ): Where, I represents the identity matrix, n(ξ) represents the normal vector, t(ξ) represents the tangent vector, and ρ(ξ), σ(ξ) respectively represent the weights of the current velocity direction affected by n(ξ), t(ξ). The expressions of ρ(ξ), σ(ξ) are as follows ρ in formula (2) 0 , σ 0 are called the repulsive force coefficient and the tangential coefficient in the obstacle avoidance principle. During the fiber reconstruction process, these two parameters actually control the influence of the tumor morphological characteristics on the selection of the fiber orientation direction. The satisfaction function τ controls the tangential influence within a set threshold, and the satisfaction function is expressed as: where g is a given positive constant, is the positive threshold, and v(ξ) is the direction of the current voxel peak image; When approaching the tumor, set the original direction as the direction from the current position to the ending area. This setting ensures that it can be tracked to the ending area. The finally perturbed fiber direction is expressed as: v 2 (ξ) = P(ξ)v(ξ) (4) Select a fixed step size according to this direction for the next step of tracking, and calculate the direction of each step until reaching the ending area; In the third step, the process of dynamically optimizing the interference matrix is: First, clustering operations are performed on the directions of the voxel peak images in the 26-neighborhood around the current position, and the results are sorted in descending order. Then, the directions corresponding to the voxels closest to the current position in each of the top 3 types of directions are taken, and this direction is regarded as the target direction to be approached. Secondly, by optimizing the repulsive force parameter, tangential direction weight ρ 0 , σ 0 in the interference matrix and satisfying the function coefficient g, the estimated direction V new after being corrected by the interference matrix and the target direction V dir have the smallest difference, that is: D(ξ) represents the distance between the current position and the center of the obstacle, M(ξ) represents the distance from the center of the obstacle to the surface at the current position, and thr represents the distance threshold set according to the tumor shape and size; When the nerve fiber tracking is far away from the tumor, it is not corrected by the obstacle equation; when approaching the tumor, regard the tumor as an obstacle and model it. The tumor is represented by an equation. Let F(ξ) be: Among them, (x 0 , y 0 , z 0 ) represents the coordinates of the tumor center, and a, b, c, d, e, f are coefficients respectively. Let F(ξ) = C, and the obstacle model is constructed by fitting the surface through the polynomial approximation algorithm.

Images