Method for automatically measuring inner ear bone labyrinth quantization parameters from ultrahigh-resolution temporal bone CT (Computed Tomography) image

Abstract

The invention discloses a method for automatically measuring inner ear bone labyrinth quantization parameters from an ultrahigh-resolution temporal bone CT image, and belongs to the field of medical image processing. According to the algorithm, anatomical structure prior information is fully utilized as guidance, a pseudo-tag proxy task is designed, and the segmentation of the inner ear bone labyrinth substructure is finally realized by combining the relevance of the whole and substructure segmentation. The inner ear bone cavity integral segmentation module adopts a segmentation network structure, and the sub-structure pseudo-tag generation module realizes geometric morphology analysis of an inner ear bone cavity structure and performs sub-structure division on an inner ear bone cavity labyrinth structure, so that sub-structure pseudo-tags are automatically generated. A single-target output layer is changed into a plurality of substructure target output layers so as to adapt to task requirements of substructure segmentation. And finally, evaluating the quality of a segmentation result through statistics of voxel distribution of each substructure, so that under the framework of active learning, the performance of a student network model can be continuously optimized, and continuous improvement of inner ear bone lost substructure segmentation is realized.

Description

Technical Field

[0001] The invention belongs to the field of image processing, and in particular relates to a method for automatically measuring quantitative parameters of inner ear bony labyrinth from ultra-high resolution temporal bone CT images. Background Art

[0002] The inner ear is the auditory and balance receptor of the human body, with a complex structure and important functions. The bony labyrinth of the inner ear includes three semicircular canals, the vestibule and the cochlea, including the bony labyrinth and the membranous labyrinth. CT imaging is an important means of examining the bony labyrinth of the inner ear.

[0003] Since the ear area is mainly composed of structures such as bone cavities and air cavities, CT images are the preferred way to intuitively display their structures. Traditional ear CT analysis and quantitative measurement are mainly done manually, which is cumbersome and results in large data differences. CT images are mainly viewed in continuous two-dimensional tomograms. In order to perform three-dimensional data analysis, the two-dimensional data obtained by computer tomography can be converted into 3D models through three-dimensional modeling, anatomical experiments, etc., to present the bone and soft tissue structures of the outer ear, middle ear, and inner ear, so as to increase the understanding of the anatomical structure of the human inner ear and middle ear, while focusing on observing specific structures and quantitatively measuring the distance and angle of the target anatomical structure.

[0004] The structure of the bony labyrinth of the inner ear is complex, the quantitative indicators that can be measured manually are limited, the measurement efficiency is low, it is highly dependent on professional experience, and the data consistency is poor.

[0005] It can be seen that by designing a method for automatic quantitative analysis of the key structures of the temporal bone, we can improve the quantitative analysis technology of ear structures and promote the understanding of the anatomical structure of the inner ear.

[0006] At present, the quantitative research work on temporal bones mostly uses relatively basic software tools for structural reconstruction and manual measurement. For the cochlea, the existing method of measuring the length of small cochlea by comparing the application of three-dimensional reconstruction technology and Archimedean spiral model uses Micro CT to scan temporal bone specimens, imports the obtained images into Mimics software to reconstruct the three-dimensional structure of the inner ear, and directly measures the length L of the cochlea, the diameter R of each turn, the height h of the cochlea, and the angle of the top turn through its built-in measurement toolkit. At the same time, the Archimedean spiral model is used to indirectly calculate the formula for the length L of the cochlea, and the difference in the length of the cochlea obtained by the two methods is compared. The three-dimensional reconstruction technology truly reflects the three-dimensional structure of the inner ear and allows direct measurement, and the measurement results are accurate and reliable. Although Micro CT has the characteristics of high image resolution, its limitation is that the field of view is small. For the semicircular canals, the existing methods of semi-automatic segmentation of the semicircular canals, automatic extraction of the semicircular canal centerline, and calculation of the semicircular canal plane fitting equation to study the spatial posture of the semicircular canals are more reliable than manual measurement, but the disadvantage is that the head-space coordinate system is not established, so the plane equations of each semicircular canal are not included in the standard space coordinate system, and the application of the measurement results is limited. Compared with manual measurement after modeling, semi-automatic segmentation improves efficiency to a certain extent, but it cannot be widely used. At the same time, the above methods are all based on a small number of ear CT segmentation models and infer information from small data sets.

[0007] For the U-HRCT inner ear bony labyrinth, there are many types of inner ear malformations, with various definitions and standards, and limited data. Traditional manual detection is greatly influenced by subjectivity, and more advanced and objective detection methods need to be explored. Therefore, it is necessary to use algorithms for automatic quantitative analysis of key structures of the inner ear to improve the efficiency of quantitative analysis of ear structures.

[0008] Based on ultra-high resolution temporal bone CT images, the present invention takes the semicircular canals, vestibule and cochlea in the inner ear bony labyrinth structure as the main research objects, designs an automatic measurement analysis algorithm to automatically extract the quantitative parameters of the inner ear bony labyrinth structure, and realizes an intelligent, fast and accurate U-HRCT (Ultra-High Resolution CT) inner ear bony labyrinth quantitative parameter analysis system. Summary of the invention

[0009] The core technology of the present invention is to propose a method for automatically measuring the quantitative parameters of the inner ear bony labyrinth using U-HRCT, and to realize automatic quantitative analysis of the quantitative parameters of key structures. The method includes automatic segmentation of the inner ear bony labyrinth substructures and automatic measurement of the substructures.

[0010] (1) Automatic segmentation of the inner ear bony labyrinth substructures

[0011] Firstly, the substructure segmentation task is completed by adopting the inner ear bony labyrinth substructure segmentation method based on morphological pseudo-labels.

[0012] The algorithm makes full use of the anatomical structure prior information as a guide, designs pseudo-label proxy tasks, combines the correlation between the overall and substructure segmentation, designs the teacher-student architecture, and finally achieves the segmentation of the inner ear bony labyrinth substructure. Its framework specifically consists of two parts: the teacher branch and the student branch.

[0013] The teacher branch consists of an inner ear bone cavity overall segmentation module and a substructure pseudo-label generation module. The inner ear bone cavity overall segmentation module adopts the segmentation network structure in previous work, while the substructure pseudo-label generation module realizes the geometric morphology analysis of the inner ear bone cavity structure and divides the inner ear bone cavity labyrinth structure into substructures, thereby automatically generating substructure pseudo-labels.

[0014] The student branch is improved from the overall segmentation network of the inner ear bone cavity. The specific improvement is to change the single target output layer to multiple substructure target output layers to meet the task requirements of substructure segmentation. Between the cascaded dual teacher branches and the student branch, a knowledge transfer mechanism is implemented, including the backbone parameter sharing strategy and knowledge distillation strategy from the overall segmentation network to the substructure segmentation network, to promote the learning efficiency of the student network branch.

[0015] Finally, the quality of the segmentation results is evaluated by statistics on the voxel distribution of each substructure, and outlier data is screened out. In this way, under the framework of active learning, the performance of the student network model can be continuously optimized to achieve continuous improvement in the segmentation of the inner ear bony labyrinth substructures.

[0016] (2) Automatic structural measurement

[0017] On the basis of the specific implementation of automatic substructure segmentation, the present invention proposes an automatic measurement algorithm of inner ear substructure based on U-HRCT.

[0018] First, a substructure voxel model is constructed using CT scans, and then reconstructed into a 3D point cloud through image processing and 3D reconstruction algorithms. In this process, the algorithm analyzes each layer of the CT scan, identifies the boundaries and features of the substructure voxels, and then integrates this 2D information to form a complete point cloud data in 3D space.

[0019] Secondly, an automatic fitting measurement algorithm was proposed for the voxel coordinate data of the key structures of the inner ear bony labyrinth. The specific measurement indicators include: semicircular canal curvature, length, plane angle, vestibular volume, cochlear volume, cochlear spiral length and other indicators.

[0020] The specific parameter measurement steps include:

[0021] (a) Extraction of the semicircular canal centerline by binary thinning algorithm;

[0022] (b) fitting it into a three-dimensional curve model using the least squares method;

[0023] (c) Automatically measure the spatial attitude, curvature and other parameters of the semicircular canals based on the fitting curve. The spatial attitude parameters specifically include the normal vectors of the three semicircular canal fitting planes;

[0024] (d) Fitting the inner ear vestibule segmentation template into an ellipsoid model and measuring its quantitative parameters such as volume and cross-sectional area;

[0025] (e) Calculate the volume of the cochlea based on the cochlear segmentation results;

[0026] (f) performing principal component analysis on the spatial coordinate data of the target voxels of the cochlear segmentation to obtain the principal component feature vector; performing structural rotation on the cochlea with the three principal component vectors as coordinates to calculate the length, width, and height of the cochlea;

[0027] (g) The coordinates of the cochlear apex are calculated by retrieving the voxel at the highest point of the cochlea, and the perpendicular line through the cochlear apex is defined as the estimated direction of the modiolar axis.

[0028] Finally, by comparing the data of U-HRCT samples of various types of inner ear bony labyrinthine malformations with normal samples, the algorithm can batch calculate the statistical information such as the mean value and standard deviation of these key indicators.

[0029] The present invention is achieved by adopting the following technical means:

[0030] A method for automatically measuring the fitting parameters of the bony labyrinth of the inner ear.

[0031] The method is divided into three parts as a whole: (automatic segmentation and calibration of the inner ear bony labyrinth substructures) reading in and three-dimensional visualization of the voxel data of the inner ear bony labyrinth substructures, automatic measurement algorithm design and program implementation of quantitative indicators of the inner ear bone cavity, and indicator testing and statistical result analysis of batch samples.

[0032] The method specifically comprises the following steps:

[0033] 1. Import and 3D visualization of voxel data of the inner ear bony labyrinth substructure

[0034] The first part of the present invention is the computer reading and display of inner ear segmentation data, using MATLAB, which has strong operability and rich function toolkit, to read and display CT image nii files and their annotation files. The idea is to study and apply nii function library, use load_nii to view nii type image files and modify pixel data.

[0035] The specific implementation steps are to first import the corresponding data package, use the corresponding algorithm to read the nii image data, display the point cloud in the window position, and finally save and visualize the results.

[0036] 2. Design of automatic measurement algorithm and implementation of program for quantitative indexes of inner ear bone cavity

[0037] This algorithm is mainly designed around quantitative parameters related to the semicircular canals, vestibule, and cochlea, including inner ear bone cavity volume calculation, semicircular canal center curve extraction, semicircular canal length and curvature calculation, inner ear vestibule ellipsoid fitting, cochlear volume, cochlear spiral length and other indicators.

[0038] ① Fitting and measurement of semicircular canal structure

[0039] The semicircular canals are semicircular bone canals, which are divided into the anterior, superior and posterior semicircular canals, presenting a certain angle and a C-shape on an appropriate plane. To achieve the measurement of parameters related to the semicircular canals, the core technologies include centerline extraction, plane fitting estimation and calculation of spatial angles. The vestibule of the inner ear mainly includes the ellipsoid saccule and the saccule, and is also connected to the three semicircular canals.

[0040] The first step is to use the algorithm to extract the 3D skeleton of the semicircular canal for human-computer interaction. The idea is to use topological correlation function programming. The algorithm mainly uses the binary thinning method, which mainly finds the center line (skeleton) of the object in the input image through a decision tree. The general method is to iteratively erode the surface of the object until only the skeleton remains. The erosion will be performed symmetrically to ensure the medial position of the skeleton line and retain the connectivity of the object without generating holes or cavities on the object. The semicircular canal structure voxels obtained by U-HRCT scanning are iteratively eroded to obtain the coordinates of each point on the center line of the semicircular canal for fitting processing.

[0041] In the second step, curve fitting is performed based on the extracted semicircular canal centerline coordinates. The algorithm automatically fits the centerline point set corresponding to each sample using the mathematical least squares curve fitting principle and saves the results.

[0042] The core principle of the least squares method is to find z∈Φ in a given function class Φ for a given data point so that the sum of the squares of the errors E 2 Minimum. In geometric terms, it is to find the minimum value that is equal to the given point {(X i , Y i , Z i )}(i=0,1,…,m) is the curve z with the smallest sum of squares of distances. Function z is called fitting function or least square solution, and the method of finding fitting function z is called least square method of curve fitting. According to the geometric characteristics of the semicircular canals, the algorithm determines that there are mainly three parameters in its fitting curve model.

[0043] The semicircular canal model is defined as: y = ax 3 +bx 2+cx+d. Therefore, the idea of ​​this step is to construct an analytical function (the semicircular canal is a three-dimensional curve model) based on the known data set at discrete points, that is, when the function value on the point set is known, so as to make the value as close to the given value as possible at the original discrete points, so as to obtain the centerline fitting curve and calculate the curvature and other data.

[0044] First, the original bone line data is determined based on the x, y, and z data, and the midpoint coordinates are obtained after integer processing. Then the maximum curvature is calculated and the maximum curvature point is marked. Finally, the fitting plane equation and its coefficients and normal vector are obtained to calculate the angle between the fitting plane and the sagittal plane, coronal plane, horizontal plane, and the other two semicircular canals in the same sample.

[0045] Finally, to quantify the volume, the algorithm automatically counts the number of voxels and calculates the physical volume using the CT resolution.

[0046] ② Fitting and automatic measurement of vestibular structure

[0047] The vestibule can be fitted as an ellipsoidal model for algorithm design because its shape is approximately ellipsoidal. When fitting the ellipsoid of the vestibular voxel point set and performing statistical analysis of the fitting data, three key steps are mainly followed:

[0048] The first step is to preprocess the data and construct the ellipsoid equation. First, the vestibular voxel point set is extracted from the original data. These point sets represent the discrete sampling of the ellipsoid surface to be fitted. Then, the basic equation form of the ellipsoid (i.e. (x / a) 2 +(y / b) 2 +(z / c) 2 =1, where a, b, and c are the three-axis radii of the ellipsoid) into the form of a general linear equation. This transformation is based on algebraic transformation and introduces new variables (such as X = x 2 , Y=y 2 , Z=z 2 ) and constant terms (such as A = 1 / a 2 , B=1 / b 2 , C = 1 / c 2 ), the original ellipsoid equation can be rewritten as a linear combination form AX+BY+CZ=1, with additional quadratic terms and cross terms to take into account the directionality of the ellipsoid (such as D xy , E xz , F yz etc.), thereby constructing a system of linear equations with multiple unknown coefficients.

[0049] The second step is to construct a linear system of equations and fit it using the least squares method. Next, the pre-processed voxel point set is substituted into the transformed linear system of equations. Each point contributes an equation to the system, forming an overdetermined system of equations (i.e., a system of equations in which the number of equations is far greater than the number of unknowns). In order to solve this system of equations, the present invention adopts the least squares method, which aims to find a set of parameter values ​​so that the residual sum of squares of all equations is minimized. Specifically, by constructing a design matrix, which contains the coordinate squares and cross terms of all points, and a response vector (usually a constant term or a linear combination), mathematical software or an algorithm is then used to solve the linear least squares problem to obtain the coefficients (i.e., A, B, C, etc.) in the ellipsoid equation.

[0050] The third step is parameter analysis and result display. Finally, the three-axis radius a, b, and c of the ellipsoid are inversely solved from the solved coefficients (calculated by a=1 / sqrt(A) and so on), and the mean and variance of these radii are calculated to evaluate the central tendency and dispersion of the data. In addition, the fitting effect of the vestibular ellipsoid is intuitively displayed using a visualization algorithm to verify the rationality of the fitting.

[0051] ③Cochlear part

[0052] The shape of the cochlea is similar to a spiral structure. The PCA principal component analysis method can be used to calibrate and reduce the dimension of the cochlear data space and simplify the measurement.

[0053] The first step is to retrieve all the voxel coordinates of the cochlea as coords1, count the number of coordinates, and convert them into volume units through CT sampling resolution. Perform PCA principal component analysis on the voxel coordinates of the cochlea, and perform dimensionality reduction and spatial position calibration on the voxel coordinates of the cochlea. The data reduced to two dimensions presents the largest projection of the cochlea, and after the spatial position, the cochlea presents a spiral shape along the Z axis, that is, the modiolus is parallel to the Z axis. The coordinates reduced to two dimensions are coords2, and the coordinates calibrated in spatial position are coords3;

[0054] The second step is to retrieve the coordinate (x1, y1) with the largest x coordinate in the two-dimensional coordinate coords2, which is the center of the round window mold. Calculate the Euclidean distance of all points from the center of the round window mold, and find the point with the farthest distance (x2, y2). At this time, the farthest distance is the base long diameter α of the cochlea; establish a new coordinate system with the straight line where the two measurement points of the base of the cochlea are located as the X-axis. In the new coordinate system, group all the coordinates with the same x coordinate into a group, traverse all groups, and locate the two points (x3, y3) and (x4, y4) with the largest difference in the Y-axis coordinate values ​​of the two coordinates. |y3-y4| is the measured short diameter β of the cochlea; the cochlear duct length CDL = 4.16α-4 is deduced from the calculated base long diameter α of the cochlea; the cochlear duct two-turn length 2TL = 3.65 (α-1); the cochlear base turn length BTL = 2.43 (α-1);

[0055] The third step is to retrieve the points with the largest and smallest absolute values ​​of the z coordinates in the coordinates coords3 of the spatial position calibration. The coordinates of Min|Z| (x3, y3, z3) are the apex of the cochlea, and the coordinates of Max|Z| (x4, y4, z4) are the base of the cochlea. Cochlear height H = (max|Z|-min|Z|);

[0056] The fourth step is to obtain the third principal component vector L3 of the spatial position calibration, which is the direction of the modiolus. The cosine values ​​C1, C2, and C3 of the angles between the modiolus and the YOZ plane, XOZ plane, and XOY plane are obtained by multiplying [1, 0, 0], [0, 1, 0], and [0, 0, 1] by the modulus length |L3| of the modiolus. The arccos function is used to obtain the radian corresponding to the cosine value and convert it into angles θ1, θ2, and θ3.

[0057] 3. Index testing and statistical results analysis of batch samples

[0058] For the index test and statistical plan analysis of batch samples, based on the completion of the first two items, the rich statistical functions of MATLAB are used to perform statistical mean values, variances and other data of large-scale quantitative indicators, and visualized using histograms, etc. to achieve measurement functions.

[0059] After obtaining the key parameters, the algorithm automatically stores the values ​​in a table and automatically performs statistics after completing the measurement of all samples. This step is mainly divided into two parts.

[0060] The first part is data preprocessing. After loading the data, necessary data cleaning and preprocessing steps are performed to detect and remove outliers to ensure the accuracy and consistency of the data.

[0061] The second part is the statistical calculation of quantitative indicators. The statistical function library of MATLAB is used to perform statistical calculations of quantitative indicators on the cleaned data. The main focus is on calculating the mean value, standard deviation and other statistics of each indicator. These statistics can reflect the central trend, dispersion and distribution of the data, and make graphs to show the data distribution more intuitively.

[0062] Compared with the prior art, the present invention has the following obvious advantages and beneficial effects:

[0063] 1. Automatic quantitative analysis of key structures of the inner ear

[0064] The present invention will design a special evaluation algorithm to automatically estimate the parameters such as the semicircular canal curvature, the angle between the three semicircular canal planes, the length and number of turns of the cochlear centerline, the volume of the inner ear bone cavity, and the volume of the vestibule, and will generate a quantitative index automatic statistical software tool.

[0065] 2. Automatic statistical analysis of large samples

[0066] The present invention uses automatic quantitative analysis technology to perform indicator statistical analysis on large sample (about 500-1000 cases) data. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 , Overall diagram of the method for automatically measuring the quantitative parameters of the inner ear bony labyrinth

[0068] Figure 2 , pseudo-label generation module framework diagram

[0069] Figure 3 , Semicircular canal measurement framework diagram

[0070] Figure 4 , Schematic diagram of vestibular ellipsoid fitting structure

[0071] Figure 5 , Visualization of cochlear structure and its template comparison

[0072] Figure 6 , indicator batch statistical variance diagram and histogram DETAILED DESCRIPTION

[0073] The following is a description of the implementation examples of the present invention in conjunction with the accompanying drawings:

[0074] The present invention aims to achieve accurate index measurement and data statistics of NIfTI (Neuroimaging Informatics Technology Initiative) format data of temporal bone U-HRCT.

[0075] A complete scan result is composed of bilateral ear data, and the unilateral data contains 370 images axially, with a scanning receptive field of 65mm×65mm×37mm, a spatial resolution of 650×650×370, a layer thickness of 0.1mm, a layer spacing of 0.1mm, a tube voltage of 100KV, and a tube current of 3.5mA.

[0076] For each U-HRCT data set, the process of automatically measuring the quantitative parameters of the inner ear bony labyrinth is as follows:

[0077] The first step, such as Figure 2 As shown in the figure, the inner ear bony labyrinth substructure segmentation method based on morphological pseudo-labeling is used to complete the substructure segmentation task, and each complete scan result is segmented into five substructures. Guided by the anatomical structure prior information, the pseudo-labeling proxy task is designed, and the correlation between the overall and substructure segmentation is combined to design a teacher-student architecture to achieve the segmentation of the inner ear bony labyrinth substructures and obtain the segmentation results, namely the cochlea, three semicircular canals and vestibule.

[0078] The second step is, Figure 3 As shown in Figure 1, the center lines of the three semicircular canals are extracted and their parameters are measured. First, the center lines of the semicircular canals are extracted using the binary thinning principle, as shown in Figure 1. Figure 3 (b) As shown in Figure 2. Secondly, the center line point set is fitted into a three-dimensional curve model by the least squares method, so as to obtain its plane equation and maximum curvature and other indicators, such as Figure 3 (c) As shown in the figure. The normal vector is obtained through the plane equation, and then the angle between the fitting plane and the sagittal plane, coronal plane, horizontal plane, and the other two semicircular canals in the same sample is obtained. Finally, the algorithm automatically counts the number of voxel points and calculates the physical volume through the resolution of CT.

[0079] In the third step, the vestibular point set is fitted into an ellipsoid model using the least squares method to measure its quantitative parameters such as volume and cross-sectional area, such as Figure 4 shown.

[0080] The fourth step is Figure 5 As shown in the figure, based on the segmentation results of the cochlear structure, the voxel coordinates of the cochlea are retrieved, the volume is counted and converted, and PCA dimension reduction and spatial calibration are performed to obtain the two-dimensional coordinates coords2 and the spatial calibration coordinates coords3. The center of the round window model and the long diameter α of the cochlear base are determined in the two-dimensional coordinates, the length of the cochlear duct and other parameters are calculated, and the apex and base of the cochlea are determined in the spatial calibration coordinates, and the cochlear height H is calculated. Finally, the modiolar axis direction vector L3 is obtained, and the cosine value of the angle between it and the plane of each coordinate axis is calculated by dot multiplication, and then converted into an angle to obtain specific indicator data.

[0081] The fifth step is to perform batch calculations on the data indicators through algorithms to obtain statistical information such as the average value and standard deviation of these key indicators, such as Figure 6 shown.

Claims

1. A method for automatically measuring quantitative parameters of the bony labyrinth of the inner ear from ultra-high resolution temporal bone CT images, characterized in that: The following steps are involved: 1) Reading and 3D visualization of voxel data of the inner ear bony labyrinth substructure Use MATLAB to read and display the CT image nii file and its annotation file; 2) Design of automatic measurement algorithm and implementation of program for quantitative indexes of inner ear bone cavity (1) Fitting and measurement of semicircular canal structure The semicircular canal is a semicircular bone canal, which is divided into the anterior semicircular canal, the superior semicircular canal and the posterior semicircular canal. The core technologies include centerline extraction, plane fitting estimation and spatial angle calculation. The vestibule of the inner ear includes the ellipsoid saccule and the saccule, and is also connected to the three semicircular canals. The first step is to use an algorithm to extract the 3D skeleton of the semicircular canal in human-computer interaction. The algorithm is a binary thinning method that uses a decision tree to find the center line of the object in the input image, i.e., the skeleton; it iteratively erodes the surface of the object until only the skeleton remains; the erosion will be performed symmetrically; it iteratively erodes the voxels of the semicircular canal structure obtained by U-HRCT scanning to obtain the coordinates of each point on the center line of the semicircular canal; The second step is to perform curve fitting based on the extracted semicircular canal centerline coordinates. The centerline point set corresponding to each sample is automatically fitted using the mathematical least squares curve fitting principle and the results are saved. The core principle of the least squares method is to find z∈Φ in a given function class Φ for a given data point so that the sum of the squares of the errors E 2 minimum; in geometric terms, it is to find the minimum value with respect to a given point {(X i , Y i , Z i )}(i=0,1,…,m) is the curve z with the smallest sum of squares of distances; the function z is called the fitting function or the least squares solution, and the method for finding the fitting function z is called the least squares method of curve fitting; The semicircular canal model is defined as: y = ax 3 +bx 2 +cx+d; when the data set at discrete points is known, that is, when the function value on the point set is known, an analytical function is constructed to obtain the centerline fitting curve and calculate the curvature; the physical volume is calculated by the resolution of CT; (2) Fitting and automatic measurement of vestibular structures The first step is to preprocess the data and construct the ellipsoid equation. First, the vestibular voxel point sets are extracted from the original data. These point sets represent the discrete sampling of the ellipsoid surface to be fitted. Then, the basic equation form of the ellipsoid needs to be converted into the form of a general linear equation, that is, (x / a) 2 +(y / b) 2 +(z / c) 2 =1, where a, b, and c are the three-axis radii of the ellipsoid respectively; this transformation is based on algebraic transformation, by introducing a new variable X = x 2 , Y=y 2 , Z=z 2 And the constant term A = 1 / a 2 , B=1 / b 2 , C = 1 / c 2 , the original ellipsoid equation is rewritten as a linear combination form AX+BY+CZ=1, and the addition of additional quadratic terms and cross terms constructs a system of linear equations with multiple unknown coefficients; The second step is to construct the linear equation system and fit it using the least squares method; Next, the preprocessed voxel point set is substituted into the transformed linear equations. Each point contributes an equation to the system, forming an overdetermined system of equations, i.e., a system of equations with far more equations than unknowns. The least squares method is used to minimize the sum of squares of the residuals of all equations to obtain the coefficients A, B, and C in the ellipsoid equation. The third step is parameter analysis and result display; the three-axis radius a, b, and c of the ellipsoid are inversely solved from the solved coefficients, and the mean and variance of these radii are calculated; (3) Cochlear part The first step is to retrieve all the voxel coordinates of the cochlea as coords1, count the number of coordinates, and convert them into volume units through CT sampling resolution; perform PCA principal component analysis on the voxel coordinates of the cochlea, and perform dimensionality reduction and spatial position calibration on the voxel coordinates of the cochlea; the data reduced to two dimensions presents the largest projection of the cochlea, and after the spatial position, the cochlea presents a spiral shape along the Z axis, that is, the modiolus is parallel to the Z axis; the coordinates reduced to two dimensions are coords2, and the coordinates calibrated in spatial position are coords3; The second step is to retrieve the coordinate (x1, y1) with the largest x coordinate in the two-dimensional coordinate coords2, which is the center of the circular window mold; Calculate the Euclidean distance between all points and the center of the round window model, and find the farthest point (x2, y2). At this time, the farthest distance is the base long diameter α of the cochlea. Establish a new coordinate system with the straight line where the two measurement points of the base of the cochlea are located as the X-axis. In the new coordinate system, group all the coordinates with the same x-coordinate into one group, traverse all groups, and locate the two points (x3, y3) and (x4, y4) with the largest difference in the Y-axis coordinate values ​​of the two coordinates. |y3-y4| is the measured short diameter β of the cochlea. From the calculated base long diameter α of the cochlea, the length of the cochlear duct CDL=4.16α-4; the length of the two turns of the cochlear duct 2TL=3.65(α-1); the length of the base turn of the cochlea BTL=2.43(α-1); The third step is to retrieve the points with the largest and smallest absolute values ​​of the z coordinates in the coordinates coords3 of the spatial position calibration; the coordinates of Min|Z| (x3, y3, z3) are the apex of the cochlea, and the coordinates of Max|Z| (x4, y4, z4) are the base of the cochlea; the height of the cochlea H = (max|Z|-min|Z|); The fourth step is to obtain the third principal component vector L3 of the spatial position calibration, which is the direction of the modiolus; multiply by [1, 0, 0], [0, 1, 0], [0, 0, 1] and divide by the modulus length |L3| of the modiolus to obtain the cosine values ​​C1, C2, C3 of the angles between the modiolus and the YOZ plane, XOZ plane, and XOY plane, and use the arccos function to obtain the radian corresponding to the cosine value and convert it into angles θ1, θ2, θ3; 3) Index testing and statistical results analysis of batch samples MATLAB is used to calculate the mean and variance of a large number of quantitative indicators, and histograms are used for visualization. When the key parameters have been obtained, the algorithm automatically stores the values ​​in a table and performs automatic statistics after completing the measurement of all samples.

Images