A Facial Acoustic Nerve Reconstruction Method Based on Global Trajectory Distribution Estimation

Abstract

A facial auditory nerve reconstruction method based on global trajectory distribution estimation. Set the starting region, ending region of the fibers and the region where the cisternal segment of the facial auditory nerve may pass according to the T2 image. Generate a peak image according to the preoperative diffusion magnetic resonance imaging (DWI image). Establish a flow field distribution model based on the above prior knowledge and the peak image, and then the global trajectory distribution function can be automatically generated. Finally, starting from the starting region, generate the streamline representing the facial auditory nerve automatically according to the global trajectory distribution function. The present invention realizes a high success rate of preoperative reconstruction of the facial auditory nerve, helps doctors better plan before surgery, and greatly reduces intraoperative injury.

Description

Technical Field

[0001] The present invention relates to the fields of medical imaging, image segmentation and tracking under computer graphics, and is a facial and auditory nerve reconstruction method based on global trajectory distribution estimation. Background Art

[0002] Vestibular schwannoma is a benign tumor located between the internal auditory canal and the cerebellar peduncle. Vestibular schwannoma usually compresses the facial and auditory nerves (the VIIth and VIIIth cranial nerves), resulting in their dysfunction. The main symptoms of patients with vestibular schwannoma include unilateral sensorineural deafness, vertigo attacks, and impaired facial function. The main treatment method for vestibular schwannoma is tumor resection surgery, which has a high risk of nerve injury. A large number of studies have shown that preoperative identification of the facial and auditory nerves can improve the nerve preservation rate and predict postoperative risks.

[0003] The traditional method for preoperative identification of the facial and auditory nerves is for surgeons to estimate through T1 and T2 images. However, it is very difficult to identify the facial and auditory nerves in this way in the case of large tumors because the signal values of the facial and auditory nerves and vestibular schwannoma in traditional magnetic resonance imaging are very similar.

[0004] Diffusion magnetic resonance tractography is a potential tool for preoperative prediction and identification of the facial and auditory nerves. The commonly used algorithms for identifying the facial and auditory nerves through diffusion magnetic resonance tractography are deterministic fiber tracking algorithms and probabilistic fiber tracking algorithms. The fiber propagation direction of the deterministic tracking algorithm is uniquely determined, that is, the direction with the highest diffusion signal value. Therefore, the ability of the deterministic tracking algorithm to suppress noise is limited. The probabilistic tracking algorithm can make up for some of the defects of the deterministic tracking algorithm. The probabilistic tracking algorithm can overcome some noise and perform better in the presence of tumors. Because the fiber propagation direction of the probabilistic tracking algorithm describes the microscopic structure information of the fibers in a probabilistic form. However, there are a large number of false positive fibers in the results of the probabilistic tracking algorithm. Simply put, the reconstruction results of the facial and auditory nerves by the probabilistic tracking algorithm may have many paths from the internal auditory canal to the cerebellar peduncle.

[0005] It can be seen from the above methods that the biggest problem faced by the current fiber reconstruction method for the facial and auditory nerves under tumor compression is partial volume effect, head motion noise, image noise, and the complex spatial structure of the facial and auditory nerves after being compressed by tumors. Summary of the Invention

[0006] In order to overcome the deficiencies of the prior art, the present invention provides a facial and auditory nerve reconstruction method based on global trajectory distribution estimation, which considers partial volume effect, head motion noise and image noise from a global perspective to solve the problem of cumulative error in fiber tracking algorithms based on local direction information.

[0007] The technical solution adopted by the present invention to solve its technical problems is as follows:

[0008] A facial and auditory nerve reconstruction method based on global trajectory distribution estimation, comprising the following steps:

[0009] Step 1: Data preprocessing

[0010] Using the structural MRI and DTI images of patients with vestibular schwannoma, select the T2 image in the structural MRI and the peak image generated from the DTI image. Among them, the peak image represents the three directions with the strongest water molecule diffusion signal for each voxel. For vestibular schwannoma data, draw the area where the cisternal segment of the facial and auditory nerve may pass (mask image) and two regions of interest of the facial and auditory nerve, and these two regions of interest are located at the cerebellopontine angle and the internal auditory canal respectively. Among them, the mask image is obtained by inflating several voxels around the edge of the tumor. Then, denoise the DTI image, correct for eddy currents and head motion, and linearly register the T2 image using the DTI image as a template;

[0011] Step 2: Representation of prior knowledge streamlines

[0012] Using the mask image and the region-of-interest image in Step 1, obtain the fiber bundle with prior knowledge through the fiber tracking method. Obtain the average direction vector of each voxel by averaging the tangent vectors of each fiber at a voxel. The global tracking region, that is, the global mask image, can be obtained through the region passed by this fiber bundle;

[0013] Step 3: Construction of the global trajectory distribution function

[0014] First, construct the streamline differential equation, incorporate the direction information of each voxel in the global mask image, and obtain the global trajectory distribution function by minimizing the cost between the trajectory distribution and the selected direction, that is, the vector field that can describe the facial and auditory nerve pathway;

[0015] Step 4: Global tracking

[0016] Generate the three-dimensional facial and auditory nerve reconstruction result through the global trajectory distribution function in Step 3. The fiber bundle starts from the internal auditory canal, and a fixed step size is set. The tracking direction of each step is obtained through the vector field, and the fiber tracking is terminated only when the fiber reaches the cerebellar peduncle.

[0017] Furthermore, in the above-mentioned Step 2, the process of representing the prior knowledge streamlines is as follows:

[0018] Taking the mask image and the region of interest image in Step 1 as inputs, voxel direction information with prior knowledge is obtained through a fiber tracking method. The starting region of this fiber tracking method is the internal auditory canal, the ending region is the cerebellar peduncle, and a loose tracking threshold is set. Fibers are tracked and fiber bundles are generated within the mask region;

[0019] In a three-dimensional space consistent with the resolution of the dMRI image, the fiber bundles are mapped into this three-dimensional space to generate a global mask image. That is, for a certain voxel in this three-dimensional space, as long as at least one fiber bundle passes through it, the value of this voxel is set to "1", and the values of other voxels are set to "0";

[0020] Finally, the Euclidean distance between each prior fiber and the internal auditory canal is calculated respectively, the endpoint with a smaller distance is found and set as the starting point, then the other endpoint is the ending point. Thus, the fiber bundle is given a direction. Therefore, the average direction vector of this voxel is obtained by averaging the tangent vectors of each fiber at this voxel.

[0021] Further, in Step 3, the construction process of the global trajectory distribution function is as follows:

[0022] Assume that the trajectory of the white matter fiber bundle is represented as a curve flow described by a vector field, and then the global trajectory distribution function is parameterized through a set of streamlines S = {s m , m = 1,..., M}, assuming

[0023] V(x, y, z) = [V x (x, y, z), V y (x, y, z), V z (x, y, z)] T (1)

[0024] is the field vector at the position (x, y, z), where

[0025]

[0026] Assume that this field vector is represented by a field approximated by a ternary N-order function, and its form is f N (x, y, z),

[0027]

[0028] where a i,j,k is the constant coefficient of this polynomial. Combining formula (3), the field vector in formula (1) is expressed as,

[0029]

[0030] where and Respectively represent the diffusion vector values V(x, y, z) in the x, y, and z-axis directions at the position (x, y, z). C is a 1×M vector, expressed as,

[0031] C(x, y, z) = [1 z... z n ... x n = [x i y j z k i＝0,...,N；j＝0,...,N-i；k＝0,...,N-i-j (5)

[0032] A is a 3×M constant coefficient matrix composed of a i,j,k denoted as,

[0033]

[0034] where M = (N + 1)(N + 2)(N + 3) / 6, and the estimation of the constant coefficient matrix A of the global trajectory distribution function is obtained by globally optimizing the following cost function

[0035]

[0036] The present invention regards the compressed facial-acoustic nerve as a hydrodynamics problem and uses Riemannian manifold to describe the distribution of the nerve. First, the region of interest images of the internal auditory canal and the cerebellar peduncle are obtained by manual annotation for each patient. Then, with the help of fiber tractography to express this prior information, we obtain the peak image from the DWI image, and fuse the prior information obtained by this fiber tracking with the peak image to obtain the global trajectory distribution function. Finally, the three-dimensional facial-acoustic nerve is reconstructed through the global trajectory distribution function.

[0037] The beneficial effects of the present invention are as follows: The present invention sets the starting region, ending region and the possible passing region of the cisternal segment of the facial-acoustic nerve according to the T2 image. According to the preoperative diffusion magnetic resonance imaging (DWI image), the peak image is generated. According to the above prior knowledge and the peak image, a flow field distribution model is established, and then the global trajectory distribution function can be automatically generated. Finally, starting from the starting region, the streamline representing the facial-acoustic nerve is automatically generated according to the global trajectory distribution function. This method realizes a high success rate of preoperative reconstruction of the facial-acoustic nerve, helps doctors better plan before surgery, and greatly reduces intraoperative injury. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is the flow chart of the implementation steps of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0039] The present invention will be further described below.

[0040] Refer to Figure 1 ​, A facial and auditory nerve reconstruction method based on global trajectory distribution estimation, comprising the following steps:

[0041] Step 1: Data preprocessing

[0042] Prepare the structural MRI and DTI images of patients with vestibular schwannoma, and select the T2 image in the structural MRI and the peak image generated from the DTI image. Among them, the peak image represents the three directions with the strongest water molecule diffusion signal for each voxel.

[0043] For vestibular schwannoma data, draw the area where the cisternal segment of the facial and auditory nerves may pass (mask image) and two regions of interest of the facial and auditory nerves. These two regions of interest are located at the cerebellopontine angle and the internal auditory canal respectively. Among them, the mask image is obtained by inflating several voxels around the edge of the tumor; then denoise the DTI image, correct for eddy currents and head motion, and linearly register the T2 image using the DTI image as a template.

[0044] Step 2: Representation of prior knowledge streamlines

[0045] Using the mask image and the region-of-interest image from Step 1, obtain fiber bundles with prior knowledge through a fiber tracking method. Calculate the average direction vector of each voxel by averaging the tangent vectors of each fiber at a voxel. The global tracking region, that is, the global mask image, can be obtained through the region passed by this fiber bundle.

[0046] In the above Step 2, the process of representing prior knowledge streamlines is as follows:

[0047] Take the mask image and the region-of-interest image from Step 1 as inputs, and obtain the voxel direction information with prior knowledge through a fiber tracking method. The starting region of this fiber tracking method is the internal auditory canal, the ending region is the cerebellar peduncle, and a loose tracking threshold is set to track and generate fiber bundles within the mask region.

[0048] Create a three-dimensional space consistent with the resolution of the dMRI image, and map the fiber bundles into this three-dimensional space to generate a global mask image. Simply put, for a voxel in this three-dimensional space, if at least one fiber bundle passes through it, the value of this voxel is set to "1", and the values of other voxels are set to "0".

[0049] Finally, calculate the Euclidean distance between each prior fiber and the internal auditory canal respectively, find the endpoint with a smaller distance, set it as the starting point, and then the other endpoint is the ending point. Thus, the fiber bundle is given a direction. Therefore, calculate the average direction vector of each voxel by averaging the tangent vectors of each fiber at this voxel.

[0050] Step 3: Construction of the global trajectory distribution function

[0051] First, construct the streamline differential equation, incorporate the direction information of each voxel in the global mask image, and obtain the global trajectory distribution function by minimizing the cost between the trajectory distribution and the selected direction, that is, the vector field that can describe the facial auditory nerve pathway;

[0052] In the third step, the construction process of the global trajectory distribution function is as follows:

[0053] Assume that the trajectory of the white matter fiber bundle is represented as the curve flow described by the vector field, and then parameterize the global trajectory distribution function through a set of streamlines S = {s m , m = 1,..., M}, assume

[0054] V(x, y, z) = [V x (x, y, z), V y (x, y, z), V z (x, y, z)] T (8)

[0055] is the field vector at the position (x, y, z), where

[0056]

[0057] Assume that the field vector is represented by a field approximated by a ternary N-order function, and its form is f N (x, y, z),

[0058]

[0059] where a i,j,k is the constant coefficient of this polynomial. Combining formula (3), the field vector of formula (1) is expressed as,

[0060]

[0061] where and respectively represent the diffusion vector values V(x, y, z) in the x, y, and z-axis directions at the position (x, y, z). C is a 1×M vector, expressed as,

[0062] C(x, y, z) = [1 z... z n ... x n = [x i y j z k i＝0,...,N；j＝0,...,N-i；k＝0,...,N-i-j (12)

[0063] A is a 3×M constant coefficient matrix composed of a i,j,k , denoted as,​

[0064]

[0065] where M = (N + 1)(N + 2)(N + 3) / 6, and the estimation of the constant coefficient matrix A of the global trajectory distribution function is obtained by globally optimizing the following cost function

[0066]

[0067] Step Four: Global tracking

[0068] Generate the three-dimensional facial auditory nerve reconstruction result through the global trajectory distribution function in Step Three. The fiber bundle starts from the internal auditory canal, and a fixed step size is set. The tracking direction of each step is obtained through the vector field. The fiber tracking terminates only when the fiber reaches the cerebellar peduncle.

[0069] The content described in the embodiments of this specification is only a list of the implementation forms of the inventive concept and is only for illustrative purposes. The protection scope of the present invention should not be regarded as limited to the specific forms stated in this embodiment. The protection scope of the present invention also extends to equivalent technical means that can be conceived by those of ordinary skill in the art based on the inventive concept of the present invention.

Claims

1. A facial and auditory nerve reconstruction method based on global trajectory distribution estimation, characterized in that: The method comprises the following steps: Step 1: Data preprocessing Using the structural MRI and DTI images of patients with vestibular schwannoma, select the T2 image in the structural MRI and the peak image generated from the DTI image. Among them, the peak image represents the three directions with the strongest water molecule diffusion signal for each voxel. For vestibular schwannoma data, draw the area where the cisternal segment of the facial and auditory nerve may pass through and two regions of interest of the facial and auditory nerve, which are located at the cerebellopontine angle and the internal auditory canal respectively. Among them, the mask image is obtained by dilating the voxels around the edge of the tumor, and then denoise the DTI image, correct the eddy current and head movement, and linearly register the T2 image with the DTI image as the template; Step 2: Prior knowledge streamline representation Using the mask image and the region of interest image in Step 1, obtain the fiber bundle with prior knowledge through the fiber tracking method. Obtain the average direction vector of this voxel by averaging the tangent vectors of each fiber in a voxel. The global tracking region, that is, the global mask image, can be obtained through the region passed by this fiber bundle; Step 3: Construction of global trajectory distribution function First, construct a streamline differential equation, incorporate the direction information of each voxel in the global mask image, and obtain the global trajectory distribution function by minimizing the cost between the trajectory distribution and the selected direction, that is, the vector field that can describe the facial and auditory nerve pathway; Step 4: Global tracking Generate a three-dimensional facial and auditory nerve reconstruction result through the global trajectory distribution function in Step 3. The fiber bundle starts from the internal auditory canal, and a fixed step size is set. The tracking direction of each step is obtained through the vector field. The fiber tracking is terminated only when the fiber reaches the cerebellar peduncle.

2. A facial and auditory nerve reconstruction method based on global trajectory distribution estimation as described in claim 1, characterized in that: In the said Step 2, the prior knowledge streamline representation process is as follows: Take the mask image and the region of interest image in Step 1 as inputs, obtain the voxel direction information with prior knowledge through a fiber tracking method. The starting region of this fiber tracking method is the internal auditory canal, the ending region is the cerebellar peduncle, and a loose tracking threshold is set to track and generate the fiber bundle within the mask region; Create a three-dimensional space consistent with the resolution of the dMRI image, map the fiber bundle into this three-dimensional space to generate the global mask image, that is, as long as a voxel in this three-dimensional space satisfies that at least one fiber bundle passes through, the value of this voxel is set to "1", and the values of other voxels are set to "0"; Finally, calculate the Euclidean distance between each prior fiber and the internal auditory canal respectively, find the end point with a smaller distance, set it as the starting point, then the other end point is the end point, so the fiber bundle is given a direction, and thus the average direction vector of this voxel is obtained by averaging the tangent vectors of each fiber in this voxel.

Images