WO2018019202A1

WO2018019202A1 - Method and device for detecting change of structure of image

Info

Publication number: WO2018019202A1
Application number: PCT/CN2017/094076
Authority: WO
Inventors: 高嵩; 柯永欣
Original assignee: 武汉大学
Priority date: 2016-07-25
Filing date: 2017-07-24
Publication date: 2018-02-01
Also published as: CN107464235A; CN106204624A

Abstract

Disclosed in the present invention is a method for detecting a change of a structure of an image. The method comprises the following aspects: (1) obtaining an interested region of an image to be detected, and detecting image reference values in the interested region of the image to be detected; and (2) collecting statistics and performing analysis on difference significance of a target structure in different types or classifying the target structure according to the image reference values of the interested region of the image to be detected. The present invention provides a method for precisely measuring the thickness of an object; and samples are classified under the assistance of thickness parameters of the object. The method in the present invention improves the sensitivity when of a subtle change of the structure of the image is detected, and has good development and application prospects.

Description

Method and device for detecting image structure change

cross reference

The present application is hereby incorporated by reference in its entirety in its entirety in its entirety in its entirety in the the the the the the the the the the

Technical field

The present invention relates to the field of image processing, and more particularly to a method and apparatus for detecting changes in image structure.

Background technique

Two-dimensional and three-dimensional images are widely used in various fields such as scientific research and medical diagnosis. Current medical images mainly include CT (computed tomography) images, MRI (nuclear magnetic resonance) images, B-scan images, X-ray fluoroscopic images, and the like. However, the diagnosis based on medical images is mainly based on the morphological characteristics of artificial observation images, and lacks standardized automatic analysis functions. In scientific research, CT and X-ray imaging are widely used in the detection of two-dimensional and three-dimensional morphological structures of objects, but their quantitative analysis methods are still not sensitive enough.

In patients with osteoporosis, trabecular bone imaging showed significant structural changes, while in Alzheimer's disease, brain tissue imaging showed a marked reduction in brain volume, widening of the sulci, narrowing of the cerebral gyrus, and ventricle Expanding suggests that imaging-based features can aid in the diagnosis of these diseases. At present, methods for assisting diagnosis of these diseases are not sensitive enough according to changes in imaging.

Osteoporosis is a senile metabolic disease that seriously endangers human health. It is mainly characterized by decreased bone mass, microstructural degradation and increased bone fragility, which leads to an increased risk of fracture in patients. The diagnosis of osteoporosis mainly relies on the measurement of changes in bone mass and bone microstructure. Bone density measurement is a common method for assessing changes in bone mass. The instruments currently used to measure bone density mainly include dual-energy X-ray absorptiometry (DEXA), bone quantitative CT (QCT) or peripheral bone quantitative CT (pQCT). The dual energy X-ray absorptiometry (DEXA) measurement of bone mineral density (BMD) changes is currently the gold standard for the diagnosis of osteoporosis. Although DEXA can measure bone mineral density in any part of the body, it can not distinguish cortical bone density and cancellous bone density and can only measure the two-dimensional parameters of bone tissue, so it can not effectively reflect the measurement. Changes in the bone structure of the site. Quantitative CT can distinguish between cortical and cancellous bones and can measure three-dimensional parameters of bone tissue. However, its radiation dose is higher than DEXA and its reproducibility is not as good as DEXA, so it cannot be used for follow-up evaluation of small changes in bone density or microstructure. The detection accuracy of pQCT is higher than that of DEXA, but it is mainly used to detect the density of limb bones and can not be used to detect vertebral bone density. Changes in bone density do not fully reflect changes in bone microstructure, and bone fracture parameters alone are not effective in assessing fracture risk. Studies have shown that the bone mineral density of most fracture patients is not low, and fracture risk assessment tools that combine parameters such as bone density are not effective predictors of the observed fractures in the vast majority of patients. Therefore, new and more sensitive indicators may be important for the diagnosis of osteoporosis and fracture risk assessment.

Micro-CT (MicroCT) is a high-precision 3D CT imaging technology with a spatial resolution of several micrometers. Since the thickness of trabecular bone in bone tissues of small animals such as rats and mice is in the range of several tens of micrometers, conventional CT scanning imaging cannot effectively display the fine structure of bones, so MicroCT has a wide range of evaluations on bone mass and bone microstructural changes in small animals. Applications.

The trabecular bone structure is abundant in a certain area below the proximal humerus and the distal femoral epiphysis growth plate, and osteoporosis can lead to the loss of trabecular bone structure in these areas, so the proximal humerus and the distal femur are under the line. A certain area is an ideal area for analyzing microstructural changes in trabecular bone of small animals. At present, the standard analysis method for the microstructural changes of the humerus at the proximal or distal femur of small animals is to first randomly select a region of interest below the epiphyseal growth plate and then calculate the total volume of the region of interest within the entire region of interest ( TV), bone volume (BV), bone mineral content (BMC), bone volume fraction (BV/TV), bone mineral density (BMD) and other parameters, and finally statistical analysis of the calculated parameters. Due to the uneven distribution of the longitudinal axis of the trabecular bone in the region between the sacral growth plate and the middle part of the diaphysis, the approximation decreases with the increase of the distance from the sacral line, and the current standard analysis method does not consider the trabecular beam. Bone prolongs the uneven distribution of the longitudinal axis of the bone, so current standard analytical methods do not sensitively detect subtle changes in trabecular bone structure. The micro-CT image contains the layer-by-layer distribution information of the trabecular bone inside the selected analysis area, and the current standard analysis method does not make full use of this distribution information, so the statistical analysis of the layer-by-layer distribution information of the trabecular bone is fully utilized. It will help to improve the sensitivity of detecting bone microstructural changes. In order to study the changes of the trabecular bone, it is necessary to select a certain area below the epiphyseal growth plate for analysis. However, the trabecular bone is attenuated from the growth plate of the epiphyseal line, and the measured parameters are gradually attenuated without obvious change inflection points. Therefore, there is no uniform standard for the selection of the analysis area. Researchers usually choose the analysis area according to their experience or use the trial and error method to select multiple analysis areas and then fix a significant change area for reporting. This standard analysis method is not only error-prone but also cumbersome. For the same set of data, the choice of different analysis areas has a very large impact on the results, so a selection method based on the analysis of the difference between the experimental group and the control group will help to improve the detection of bone microstructural changes. Sensitivity.

The trabecular bone thickness is another important indicator reflecting the microstructural changes of the trabecular bone. However, the measurement method of the thickness of the two-dimensional or three-dimensional image has a large error compared with the true thickness of the object, and the measured thickness of the trabecular bone cannot fully reflect the severity of the osteoporosis symptom. Therefore, a more accurate measurement of trabecular bone thickness is of great significance for the diagnosis of osteoporosis.

Alzheimer's disease is a common progressive neurodegenerative disease with clinical manifestations of memory and cognitive decline, personality and behavioral changes. It seriously affects the health of the elderly and puts tremendous pressure on society. The diagnosis of Alzheimer's disease is mainly judged from the aspects of cognitive function and emotional response through the relevant assessment form, and lacks accurate quantitative diagnosis methods. In Alzheimer's disease, brain weight and brain volume are significantly reduced, the sulcal widens, the cerebral gyrus narrows, and the ventricles enlarge. Magnetic resonance imaging analysis showed that the gray matter, white matter and ventricles in the brain of Alzheimer's patients were significantly different from those of normal controls. Because the gray matter and white matter of the brain are irregularly distributed, it is very similar to trabecular bone. Therefore, the measurement of the thickness of gray matter, white matter, and ventricle in brain images such as MRI and CT will contribute to the diagnosis of Alzheimer's disease.

In summary, the existing methods for detecting image structure changes cannot sensitively reflect subtle changes in image structure.

Summary of the invention

It is an object of the present invention to provide a method of detecting subtle changes in image structure.

Another object of the present invention is to provide a method for accurately measuring the thickness of a target structure, and assisting in the classification of the object or the diagnosis of the disease by using accurate measurement of the thickness parameter of the target structure.

According to an aspect of the present invention, a method of detecting a change in an image structure is provided, comprising the steps of:

Step 1: acquiring a region of interest of each image of the object to be detected, and determining respective image parameter values inside the region of interest of each image of the object to be detected, the image parameter value including the total volume of the region of interest, the target structure volume One or more image parameter values of the target structure material content, the thickness of the target structure, the target structure volume fraction, and the target structure density;

Step 2: statistically analyzing, according to each image parameter value of each region of the image of the object to be detected, a difference saliency of each image parameter value of the region of interest of the object to be detected in different classifications or for the object to be detected sort.

Specifically, the target structure thickness in the image parameter value includes: an average thickness of the target structure, an area or a volume of the structural portion having a specific thickness in the target structure, and other parameters derived from different thickness portions of the target structure are derived by different calculation methods. The value of the parameter that comes out.

For each two-dimensional object, firstly, according to the current general image segmentation method, each pixel of the region of interest and the target structure of the object to be detected is directly determined on the two-dimensional plane image; then each image parameter value of the target structure of the object to be detected is calculated; The individual image parameter values of the target structure of the object to be detected are statistically analyzed according to the image parameter values of the target structure of the object to be detected, and the difference significance of the image parameter values in the different classifications or the object to be detected is classified.

For each three-dimensional object, first determine the region of interest and the pixel of the target structure of the object to be detected directly on the two-dimensional plane image according to the current general image segmentation method in each of its two-dimensional planes, and then follow the steps below. Select the level of the three-dimensional region of interest; its concrete implementation includes the following sub-steps:

Step A.1: measuring each image parameter value of each three-dimensional object in different groupings in its respective two-dimensional level region of interest;

Step A.2: Selecting one or more layers of two-dimensional layers from each of the three-dimensional objects, and statistically analyzing the significance test of each image parameter value of the selected two-dimensional level region of interest between different groups. value;

Step A.3: statistically analyzing the significance value of each image parameter value of the three-dimensional object in all possible two-dimensional regions of interest in different groups according to step A.2;

Step A.4: Select the minimum p value in step A.3 or the p value less than the preset threshold, and determine the region of interest of the corresponding three-dimensional object according to the selected p value.

Preferably, the target structure is determined by one or more of a maximum square filling method, a maximum circular filling method, a maximum circular or square filling method, a maximum cube filling method, a maximum sphere filling method, a maximum sphere or a cube filling method. thickness.

Preferably, when the thickness of each pixel in the target structure is measured by the maximum square filling method, the maximum circular filling method, the maximum cube filling method, or the maximum sphere filling method, the filled maximum of the pixels including the pixel and completely contained in the target structure is included. The center of the shape is at the target structure The center of the pixel is at or at the apex of the pixel.

Preferably, when the thickness of the target structure is measured by a maximum circular or square filling method, for each pixel in the target structure, the pixel obtained by the maximum square filling method and the maximum circular filling method The maximum value in the thickness is taken as the thickness of the pixel.

Preferably, when the thickness of the target structure is measured by the maximum sphere or cube filling method, for each pixel in the target structure, the maximum sphere filling method and the maximum cube filling method are obtained in the thickness of the pixel The maximum value is taken as the thickness of the pixel.

Preferably, one or more of the repeated measurement analysis using t-test, one-way ANOVA, linear regression analysis, nonlinear regression analysis, nonlinear exponential decay regression analysis, logistic regression, repeated measurement analysis of variance, and linear mixed effect model The statistical method analyzes the difference saliency of each image parameter of the target structure of the object to be detected under different classification conditions.

Preferably, the object to be detected is classified according to each image parameter of the target structure, and specifically includes: acquiring different objects of a known classification, and calculating each target structure in the scanned image of the known classified object and the object to be detected. Image parameters, the image parameters including the thickness of the known classified object and the target structure of the object to be detected; using discriminant analysis according to each image parameter value of the known classified object and the target structure of the object to be detected and the known classification situation The one to be detected is classified by one or several statistical methods in principal component analysis, factor analysis, and logistic regression.

Preferably, the classification of the object to be detected is to classify the severity of osteoporosis of the individual bone to be determined.

Preferably, the classification of the object to be detected is to classify the severity of Alzheimer's disease in the individual to be determined.

The invention provides a method for selecting an analysis region according to the distribution of each image parameter of the target structure; the invention provides a method for hierarchically calculating each image parameter of the target structure separately and according to the layered calculation parameter in each sample The distribution within the region of interest is analyzed by appropriate statistical methods such as linear regression, nonlinear regression, and repeated measures analysis of variance; the present invention proposes an osteoporosis and altz according to relevant parameters such as image structure thickness and the like. Method for assisting diagnosis of Haimo disease; the present invention improves the method for measuring the thickness of an object structure in two-dimensional and three-dimensional images by using the maximum circular filling method and the maximum spherical filling method, taking into account all possible distribution positions of the center and the center of the sphere And all the possibilities that they are integers in diameter In this case, the accuracy of the thickness measurement is improved; the present invention proposes a new method for analyzing the thickness of objects in two-dimensional and three-dimensional images using the maximum circular or square filling method and the maximum sphere or cube filling method, which reduces the thickness measurement The above improvements, which can significantly improve the accuracy of the analysis of the target object, will help to improve the sensitivity of image-dependent detection methods, such as the evaluation of bone loss and bone structure damage in osteoporosis , evaluation of brain tissue changes in neurodegenerative diseases.

DRAWINGS

1 is a flowchart of a method for detecting a change in an image structure according to an embodiment of the present invention;

2 is a graph showing the results of distribution of various analysis parameters of the distal femur trabecular bone along the longitudinal axis of the femur according to an embodiment of the present invention;

Figure 3 is a diagram showing the effect of small osteogenic molecules on BV, TV, and BMC parameters of trabecular bone using nonlinear regression exponential decay and the use of linear regression analysis of osteogenic small molecules to trabecular BV/TV and Analysis of the impact of BMD parameters;

4 is a selection analysis diagram of an analysis region in which different parameters of the trabecular bone have significant differences under different processing conditions according to an embodiment of the present invention;

5 is a graph showing the results of distribution of various analysis parameters of the trabecular bone along the longitudinal axis of the femur in the selected analysis region according to an embodiment of the present invention;

6 is a schematic diagram of a method of maximal square filling, maximum circular filling, maximum circular or square filling according to an embodiment of the present invention;

7 is a schematic diagram of a layer-by-layer calculation method for two-dimensional thickness and three-dimensional thickness according to an embodiment of the present invention;

8 is a schematic diagram of a method for calculating an average thickness of an object according to an embodiment of the present invention;

9 is a comparative analysis result of two-dimensional thickness measurement of a standard object thickness by using different thickness measurement methods according to an embodiment of the present invention;

10 is a comparative analysis result of three-dimensional thickness measurement of a standard object thickness by using different thickness measurement methods according to an embodiment of the present invention;

Figure 11 is a graph showing the effect of osteogenic small molecules on the distribution of trabecular bone thickness from the growth plate along the long axis of the femur according to an embodiment of the present invention;

Figure 12 is a graph showing the effect of osteogenic small molecules on different two-dimensional and three-dimensional thickness parameters of trabecular bone in an embodiment of the present invention;

FIG. 13 is a diagram showing the use of hierarchical clustering method for the trabecular bone containing two-dimensional thickness according to an embodiment of the present invention. Schematic diagram of the results of the analysis with the parameters;

FIG. 14 is a schematic diagram showing the results of analyzing the different parameters of the trabecular bone containing the three-dimensional thickness by the hierarchical clustering method according to the embodiment of the present invention.

detailed description

The present invention will be further described in detail with reference to the accompanying drawings and embodiments. this invention.

At present, the detection methods for the two-dimensional image or the three-dimensional image structure are not sensitive enough. The problem to be solved by the present invention is to provide a method capable of sensitively detecting subtle changes in a two-dimensional image or a three-dimensional image structure. The invention can not only analyze based on CT imaging, but also analyze structural changes of bone tissue, brain tissue, cardiovascular, lung, kidney, etc. for small animals, large animals and human bodies based on nuclear magnetic resonance images, ultrasonic images and the like. The invention is not limited to medical images, but can be applied to any image. The invention improves the accuracy of image analysis, and thus is advantageous for the diagnosis of osteoporosis, cardiovascular and cerebrovascular diseases, neurodegenerative diseases, kidneys, lungs and the like based on structural changes.

In the present invention, the digital image for analysis consists of pixels. For a two-dimensional digital image, each pixel is represented in the present invention as a square with one center and four vertices. For a three-dimensional digital image, each pixel is represented in the present invention as a cube with one center and eight vertices.

In the present invention, the thickness parameter of the object target structure includes a two-dimensional thickness parameter or a three-dimensional thickness parameter of the object. For the two-dimensional thickness parameter, firstly, the thickness of each pixel on the object is measured by the two-dimensional thickness calculation method in each layer of the two-dimensional plane, and the two-dimensional thickness parameter includes the average thickness of the object in each layer of the two-dimensional plane, at each The layer two-dimensional plane has one or several parameters of the area of the particular two-dimensional thickness portion. For the three-dimensional thickness parameter, firstly, the thickness of each pixel on the object is measured by the three-dimensional thickness calculation method in the three-dimensional volume, and the three-dimensional thickness parameter includes the average thickness of the object in the three-dimensional volume, the volume having a specific three-dimensional thickness portion in the three-dimensional volume, and each The three-dimensional average thickness of the two-dimensional plane of the layer, one or several parameters of the volume having a particular three-dimensional thickness portion in each of the two-dimensional planes. At the same time, the thickness parameter of the object target structure also includes other parameter values derived from different parameters of the target structure through different calculation methods.

In the present invention, the treatment of the experimental animals is as follows: female rats of the age of six months are selected and divided into three groups. Twelve rats in each group were administered with PBS (control group), osteogenic small molecule (experimental group) or PTH, and three months later, the femur was isolated for scanning analysis. Four-month-old female rats were divided into four groups, 12 in each group. Three groups underwent bilateral ovariectomy (OVX group), and one group underwent sham operation (Sham group). After eight weeks of surgery, OVX group was treated with PBS. (Control group), osteogenic small molecule (experimental group) or PTH injection, and Sham group was administered by PBS injection. After three months of dosing, the femur was isolated for scanning analysis. All distal femurs were scanned and scanned with a Scanco micro-CT scanner at a spatial resolution of 15 microns.

FIG. 1 is a flowchart of a method for detecting a change in an image structure according to an embodiment of the present invention. As shown in FIG. 1 , the method includes the following steps:

Step 2: statistically analyzing the difference saliency of each image parameter value of the three-dimensional region of interest of the object to be detected in different classifications according to each image parameter value of each region of the image of the object to be detected or for the to-be-detected Objects are classified.

Specifically, the image of the object to be detected in the present invention may be a two-dimensional image or a three-dimensional image, and may be various medical images. Since the three-dimensional image is composed of a plurality of two-dimensional images, in the present embodiment, the two-dimensional image or each layer of the three-dimensional image is explained as a three-dimensional image having a unit thickness. Currently, in the analysis of three-dimensional image data, for any sample, after selecting the region of interest, each measurement parameter of the object obtains only one measurement value in the region of interest. Since the distribution of individual measurement parameters of the object within the selected region of interest is not uniform, a single measurement of each measurement parameter of the object does not retain the characteristic distribution information of the parameters, so a single measurement of each measurement parameter of the object is utilized. Subtle changes in the structure of the three-dimensional image cannot be detected sensitively. The embodiment of the present invention makes full use of the layer-by-layer distribution information of each measurement parameter in the region of interest by acquiring the measured values of the respective measurement parameters in the three-dimensional image of the object, thereby facilitating detection of the object structure. Subtle changes under different conditions improve the sensitivity of detecting structural changes in objects.

Specifically, FIG. 2 is a distribution diagram of image parameters of the distal femur trabecular bone along the longitudinal axis of the femur in the embodiment of the present invention. First, use the software to automatically identify the acquired femur at each level of the 3D level. The outer boundary of the present, and then select the inner region of the medullary cavity at a specific distance from the outer boundary to the interior of the medullary cavity along the outer boundary of the femur as the region of interest for trabecular bone analysis; then the trabecular bone is distinguished by a set threshold within the region of interest Bone and medullary cavity background; finally calculate the trabecular bone volume (BV) of each layer, the total volume of the selected region of interest (TV), the bone content of the trabecular bone (BMC), and the bone volume fraction of the trabecular bone ( BV/TV), bone mineral density (BMD) of trabecular bone. The distance from the distal femoral growth plate is taken as the X-axis, and the image parameters of the trabecular bone are plotted on the Y-axis. The results showed that the bone mass (BMC), trabecular bone volume (BV), and total volume of interest (TV) of the trabecular bone in the medullary cavity were approximately exponentially decayed along the long axis of the femur from the distal femoral growth plate. Distribution, and bone volume fraction (BV/TV), bone density (BMD) and other parameters are approximately linearly attenuated along the long axis of the femur. According to the current traditional analysis method, after selecting the analysis area, each sample only obtains one measurement value for each parameter, which is equivalent to accumulating or averaging the value of the parameter at each level in the analysis area. These accumulated or averaged measurements are then grouped using statistical methods such as t-test, analysis of variance, and linear regression. This method of value loses the distribution information of the parameter inside the analysis area, and the statistical analysis results cannot sensitively detect small structural changes. In the present invention, the bone mass (BMC), trabecular bone volume (BV), total volume of interest (TV) and other parameters of the trabecular bone in the medullary cavity are approximated from the distal femur growth plate along the long axis of the femur. The exponential decay distribution, so the nonlinear exponential decay regression statistical analysis is a suitable method for analyzing these parameters. Because the bone volume fraction (BV/TV) and bone mineral density (BMD) of the trabecular bone in the medullary cavity are approximately linearly attenuated along the long axis of the femur from the distal femoral growth plate, linear regression statistical analysis is a suitable analysis. The method of these parameters. Since linear regression statistical analysis can be regarded as a special case of nonlinear regression analysis, these parameters can be analyzed either by linear regression or by linear fitting in nonlinear regression. For each sample, each measurement parameter has a measurement at each level, and the same measurement parameters for the same sample at each level are highly correlated. For such measurement parameters, repeated measurement analysis of variance or linear mixed model repeated measurement analysis is the most appropriate analysis method, in which the scan level is equivalent to the time variable. The results show that the analysis methods used in the present invention, including the results of nonlinear exponential decay regression analysis, linear regression statistical analysis and repeated measurement analysis, are more sensitive to detecting small changes in image structure than conventional analysis methods.

Specifically, FIG. 3 is a graph showing the influence of the osteogenic small molecule on the analysis parameters of the trabecular bone by exponential decay nonlinear regression and linear regression statistical analysis in the embodiment of the present invention. PBS (CTRL) or into A small femoral (TR)-treated rat femur sample was selected from the growth plate below the trabecular bone region ending 8 mm from the growth plate, using a nonlinear regression exponential decay model (M. CTRL and M. TR) Analysis of changes in BV, TV, and BMC parameters, and analysis of changes in BV/TV and BMD parameters using linear regression models (M.CTRL and M.TR). The results show that the nonlinear regression exponential decay model has a good fitting effect on the BV, TV and BMC parameters of the trabecular bone, while the linear regression model has a poor fitting effect on the BV/TV and BMD parameters. Therefore, for the trabecular bone BV, TV, BMC parameters, nonlinear regression exponential decay model is a suitable statistical analysis method. For the BV/TV and BMD parameters of the trabecular bone, if the analysis region selection range is from the beginning of the femoral growth plate to the termination of the region containing almost no trabecular bone, the nonlinear regression exponential decay model and the linear regression statistical analysis method are very large. Error.

Since both nonlinear regression and linear regression models are not good statistical methods for analyzing BV/TV or BMD parameters of trabecular bone, the same measurement parameters of the same sample at each level are highly correlated, so repeated measurement analysis is analysis. The preferred statistical method for these trabecular bone parameters. In our sample, the parameters of the trabecular bone near the femoral growth plate or the central region of the diaphysis were small in the control group (PBS) and the treatment group (TR), and the area between the growth plate and the middle part of the backbone The difference is large. However, below the growth plate of the distal femur, the total volume of the trabecular bone (TV), bone volume (BV), bone mineral content (BMC), bone volume fraction (BV/TV), bone density (BMD) and other parameters As the distance from the growth plate at the distal end of the femur gradually increases, it gradually decreases, without a clear boundary. At present, different researchers can randomly select one or more analysis areas for analysis based on their own experience or using trial and error, and finally select the most representative areas for the results report. This method is not only time-consuming and laborious, but the selected statistically significant analysis area may be due to sample measurement error, and does not necessarily represent a significant difference in trabecular bone parameters under different processing conditions.

In order to select a level in which the parameters of the trabecular bone are significantly different between the treatment group and the control group, based on the above embodiments, the present embodiment provides a method to assist in the selection of the region of interest, and the specific implementation includes The following substeps:

Step A.2: selecting a corresponding one or more layers of two-dimensional layers from each of the three-dimensional objects, and statistically analyzing each image parameter value of the selected two-dimensional layer of interest region between different groups Significance test P value;

Specifically, FIG. 4 is a selection analysis diagram of an analysis region in which various parameters of the trabecular bone have significant differences under different processing conditions according to an embodiment of the present invention. After bilateral oophorectomy rats were treated with PBS (CTRL) or osteogenic small molecules (TR), femoral samples were taken and scanned by micro-CT. Each femur sample was calculated from the growth plate layer by layer for each parameter value of BV, BMC, BV/TV and BMD. For each parameter, a two-dimensional table is created, the position of the table cell represents the selected analysis area, and the value of the table cell is between the experimental group and the control group in the analysis area represented by the table position. Statistical p-value. The cell position (x, y) represents the analysis area from the x level to the end of the y level (x is the column where the cell is located, y is the line where the cell is located), then such a table contains the scan from All results of significant differences between the experimental and control groups of the trabecular bone parameters in all possible analysis areas from the beginning of the first layer of the sample to the end of the last layer. We performed a three-dimensional map analysis of each parameter of the trabecular bone (Fig. 4A). The results showed that in this group of experiments, the parameters of the trabecular bone were only within a small part of the analysis area between the experimental group and the control group. There was a significant difference (p < 0.05). In order to facilitate the selection of analysis areas with significant differences, we classify all the cells in the table according to the set significance threshold (α=0.05), and draw a two-dimensional image, where each pixel on the image represents two A cell of a dimension table, where cells with a p-value less than or equal to 0.05 are marked in white, cells with a p-value greater than 0.05 are marked in black, and background cells are marked in gray. The coordinates of any pixel in the two-dimensional image of FIG. 4B directly correspond to the range of the selected region of interest, and the color of any pixel represents the corresponding parameter of the trabecular bone between the treatment group and the control group. Whether there is a significant difference in the analysis area. As a preference, the selection of the analysis region may select either the region corresponding to the smallest p value in all possible analysis regions or the region corresponding to any one of the white pixels. Figure 4C is a statistical distribution of BV/TV and BMD parameters of the distal femoral trabecular bone in the sham operation group and the ovariectomized group. Four-month-old rats were subjected to sham operation (control group) or bilateral oophorectomy (experimental group). From the ninth week after surgery, PBS was intraperitoneally injected for three months, and femoral samples were scanned for analysis. The results showed that at the α=0.05 level, there was a significant difference between the experimental group and the control group in all the selected areas. However, when the α=0.00001 level, the experimental group and the control group showed significant differences only in a part of the analysis area. The results show that using a two-dimensional significant difference graph containing all the analysis regions, we can not only be used to select the analysis region for the sample with only minor changes in the analysis parameters between the experimental group and the control group, but also can be used to A sample-assisted selection analysis area with significant changes between the experimental group and the control group.

The effects of osteogenic small molecules on various parameters of trabecular bone were further analyzed in the selected analysis area. Specifically, FIG. 5 is an analysis diagram of effects of osteogenic small molecules on BV, TV, BMC, BV/TV, and BMD parameters of the trabecular bone according to an embodiment of the present invention. Rat femur samples treated with PBS (CTRL) or osteogenic small molecules (TR) were selected from growth plates below 0.5 mm from the growth plate to the trabecular bone region ending 3.0 mm from the growth plate. Among them, the first column is the distribution of the trabecular bone parameters of the control group (CTRL) and the treatment group (TR) along the long axis of the femur in the selected area, and the second to fourth columns are the control group (CTRL) and the treatment group respectively. (TR) The distribution of the trabecular bone parameters or their mean (Means) along the long axis of the femur in the selected area, and the last column is the grouped histogram of the mean values of the trabecular bone parameters within the selected area. Ns: no significant difference, *: p < 0.05, **: p < 0.01. The results show that in the selected analysis area, the parameters of the trabecular bone are approximately linearly attenuated along the long axis of the femur, so linear regression is a suitable method for analyzing these parameters. Since linear regression analysis can be regarded as a special case of nonlinear regression, linear model analysis of nonlinear regression is also a suitable statistical method for analyzing these parameters. At the same time, since each parameter of each sample has a measurement value at each layer, these values represent repeated measurements of these parameters in the sample, so repeated measurement statistical analysis is also a suitable analysis method for these parameters.

Specifically, Table 1 is a table for analyzing the influence of different statistical analysis methods of the embodiments of the present invention on the statistical results of various analysis parameters of the trabecular bone between the experimental group and the control group. After the rats were treated with PBS (CTRL) or osteogenic small molecules (TR), the femur samples were taken and scanned by micro-CT. The different analysis areas below the growth plate of the distal femur of the rats were analyzed by t-test, repeated measurement analysis, nonlinear regression linear model analysis, nonlinear regression exponential decay model analysis, etc. to analyze the osteogenic small molecules to the trabecular bone. The impact of the parameters. Among them, t-test, repeated measurement analysis and nonlinear regression linear model analysis were performed to select the area from 0.5 mm to 3.0 mm below the distal femoral growth plate, while the nonlinear regression exponential decay model analysis selected the analysis range below the distal femoral growth plate. Area from 0 mm to 8.0 mm. The results show that the linear mixed effect repeated measurement analysis is suitable It is used for the analysis of various analytical parameters of trabecular bone, and this statistical analysis method is more sensitive than the t-test method. The linear model analysis of nonlinear regression is suitable for analyzing the analytical parameters of the approximate distribution of the trabecular bone in the selected part of the distal femur, and the exponential decay model analysis of the nonlinear regression is suitable for the index of the distal femur. The individual analytical parameters of the attenuated distribution of the trabecular bone were analyzed. Since linear regression and nonlinear regression analysis require the distribution of sample residuals, repeated measurement analysis is the most appropriate method for analyzing these data if the experimental data does not satisfy the analytical conditions of linear regression and nonlinear regression.

The trabecular bone thickness is another important indicator reflecting the microstructural changes of the trabecular bone. At present, Scanco's micro-CT standard analysis software and Bonej have used the largest sphere filling method to calculate the three-dimensional trabecular bone thickness. The method of calculating the thickness of a two-dimensional structure using the maximum circular filling method was reported by Garrahan et al. in 1987. Hildebrand et al. extended the method to a situation independent of any model in 1997 and performed two-dimensional and three-dimensional thickness calculations. Due to the three-dimensional thickness gauge

Table 1 Effect of different statistical methods on statistical results of various image parameters of trabecular bone

a: repeated measurement analysis of linear mixed benefit model;

b: nonlinear regression linear model analysis;

c: Nonlinear regression exponential decay model analysis.

Calculating a large amount of computational resources, various three-dimensional trabecular bone thickness calculation methods have adopted different approximate optimizations, resulting in different algorithms to achieve a completely different thickness calculation results for the same set of three-dimensional data. In the maximum circular or maximum sphere filling process, the current software implementation does not consider all possible distributions of the center or the center of the sphere, resulting in errors in the measurement of the thickness of some objects and the actual thickness of the object. For example, for the largest circular or maximum sphere fill, if the filled circle or sphere has an odd diameter, the center or center of the sphere is at the center of one of the pixels of the object to be measured. However, if the filled circle or sphere has an even diameter, the center of the circle or the center of the sphere is located at the apex of one of the pixels of the object to be measured. Therefore, the algorithm implementation of thickness measurement needs to consider the most filled Different cases of large circle or maximum sphere center distribution, however, all current algorithm implementations do not take into account the position of all possible distributions of the center or center of the sphere, resulting in an error between the thickness measurement and the true value. In the present invention, we did not take any approximate optimization but calculated the two-dimensional or three-dimensional trabecular bone thickness directly according to the maximum circular (2D) or maximum sphere (3D) filling algorithm, and we considered the largest circular shape of the filling. The center of the sphere or the sphere of the largest sphere is distributed throughout the sphere. For standard structures with a fixed thickness, our thickness measurement method significantly improves measurement accuracy compared to conventional methods.

In a two-dimensional plane, a square-shaped object with a maximum square filling method is more accurate than a maximum circular filling method, and a circular-shaped object with a maximum circular filling method has a larger thickness measurement than a maximum square filling method. accurate. For an object of any unknown structure in a two-dimensional plane, we can see that it is composed of various square structures and various circular structures. Therefore, the thickness of the object is theoretically superior by using the maximum circular and maximum square filling methods. The thickness is measured using the largest square alone or with the largest circular filling method alone. Similarly, in the three-dimensional plane, the thickness of any three-dimensional structure measured by the maximum sphere and the maximum cube filling method is theoretically superior to the thickness measurement by using the largest sphere alone or by using the maximum cube filling method alone.

6A is a schematic diagram of a method of maximal square filling, maximum circular filling, maximum circular or square filling according to an embodiment of the present invention, which is a structure to be measured thickness from left to right, a maximum circular filling, a maximum square filling, and a maximum circular shape. Or a square fill diagram. Fig. 6B is a view showing the position of the center of the largest circular shape of the filling when the maximum circular diameter of the filling is odd and even, respectively, according to an embodiment of the present invention. Here, the maximum circular diameter of the filling is 5 pixels (odd number) or 6 pixels (even number), respectively. The results show that when the maximum circular diameter of the filling is odd, the center of the largest circle is located at the center of the positive center pixel. When the maximum circular diameter of the filling is even, the center of the largest circle is located at the intersection of 4 pixels in the center. Up, that is, at a vertex of the center pixel.

On the basis of any of the above embodiments, the present embodiment provides a method for measuring the thickness of a target structure, wherein the thickness of each pixel of the target structure and the average thickness of the target structure in the two-dimensional plane pass the maximum square filling method, and the maximum One or more of the circular filling method, the maximum square or the circular filling method are calculated.

Based on any of the above embodiments, the embodiment provides a measurement of the thickness of the target structure. The method is characterized in that the thickness of each pixel of the target structure and the average thickness of the target structure in the three-dimensional region are calculated by one or more of a maximum cube filling method, a maximum sphere filling method, a maximum cube or a sphere filling method.

Based on any of the above embodiments, the present embodiment provides a method for measuring the thickness of a target structure by using a maximum square filling method, the method comprising the following substeps: Step B.1: Image to be measured, for each pixel Calculating a maximum square containing the pixel centered on the center of any pixel and containing no background pixels, and setting the side length of the largest square to the thickness of the pixel; step B.2: the image to be measured, for each pixel Calculating a maximum square containing the pixel centered on any pixel vertex and not including the background pixel, and setting the side length of the largest square to the thickness of the pixel; step B.3: calculating the image for each pixel to be measured The thickness is the maximum of the two values of the pixel thickness calculated in steps B.1 and B.2.

Based on any of the above embodiments, the present embodiment provides a method for measuring the thickness of a target structure using a maximum circular filling method, the method comprising the following substeps: Step C.1: Image to be measured, for each The pixel calculation includes the pixel centered on the center of any pixel and does not contain the maximum circle of the background pixel, and sets the diameter of the largest circle to be the thickness of the pixel; step C.2: the image to be measured, for each One pixel calculates the largest circle containing the pixel centered on any pixel vertices and does not contain the background pixel, and sets the diameter of the largest circle to be the thickness of the pixel; step C.3: the image to be measured, calculates each The thickness of one pixel is the maximum of two values of the thickness of the pixel calculated in steps C.1 and C.2.

Based on any of the above embodiments, the present embodiment provides a method for measuring the thickness of a target structure using a maximum circular or square filling method, the method comprising the following substeps: Step D.1: Measuring by a maximum square filling method The maximum thickness of each point of the object pixel on the two-dimensional image; step D.2: measuring the maximum thickness of each point of the object pixel on the two-dimensional image by using the maximum circular filling method; step D.3: calculating each pixel of the image to be measured The thickness is the maximum of the two values of the pixel thickness calculated in steps D.1 and D.2.

Based on any of the above embodiments, the present embodiment provides a method for measuring the thickness of a target structure using a maximum cube filling method, the method comprising the following substeps: Step E.1: Image to be measured, for each pixel Calculating a maximum cube containing the pixel centered on the center of any pixel and containing no background pixels, and setting the side length of the largest cube to the thickness of the pixel; step E.2: the image to be measured, for each pixel Calculate at any pixel vertex The heart contains the pixel and does not contain the largest cube of the background pixel, and sets the side length of the largest cube to the thickness of the pixel; step E.3: the image to be measured, calculates the thickness of each pixel as step E.1 and The maximum of the two values of the pixel thickness calculated in step E.2.

Based on any of the above embodiments, the present embodiment provides a method for measuring the thickness of a target structure by using a maximum sphere filling method, the method comprising the following substeps: Step F.1: Image to be measured, for each pixel Calculating a maximum sphere containing the pixel centered on the center of any pixel and containing no background pixels, and setting the diameter of the largest sphere to the thickness of the pixel; step F.2: calculating the image to be determined for each pixel The largest sphere containing the pixel and having no background pixel as the center of any pixel vertex, and setting the diameter of the largest sphere to be the thickness of the pixel; Step F.3: calculating the thickness of each pixel for the image to be measured is The maximum of the two values of the pixel thickness calculated in steps F.1 and F.2.

Based on any of the above embodiments, the present embodiment provides a method for measuring the thickness of a target structure using a maximum sphere or cube filling method, the method comprising the following substeps: Step G.1: Measuring a three-dimensional shape using a maximum cube filling method The maximum thickness of each point of the object pixel on the image; step G.2: measuring the maximum thickness of each point of the object pixel on the three-dimensional image by using the maximum sphere filling method; step G.3: calculating the thickness of each pixel as the image to be measured The maximum of the two values of the pixel thickness calculated in G.1 and G.2.

In the process of calculating the maximum circle, the largest square, the largest sphere, or the largest cube centering on the center or vertex of any pixel and not including the background pixel for each pixel on the object target structure (the following square, circle) The sphere or cube is called a standard figure, and the largest figure of the fill is called the largest standard figure. The maximum thickness of each pixel on the object target structure can be calculated as follows: In order to calculate the maximum of any pixel p on the object target structure Thickness, the center of a pixel on a selected object or the apex of a pixel as the center of the filled standard pattern, with a radius of 1 pixel gradually increasing the standard pattern in the object until the filled standard pattern contains background pixels inside and terminates. The largest graphics that do not contain background pixels filled with this method are the largest standard graphics. If the pixel p is inside the filled maximum standard pattern, and the thickness of the pixel p is smaller than the thickness (edge length or diameter) of the maximum standard pattern, the thickness of the pixel p is updated to the thickness of the largest standard pattern. According to the above method, the maximum standard pattern is filled with the center of each pixel on the object or the vertices of the pixel, and the thickness value of the pixel p is updated. The thickness value of the pixel p obtained last is the maximum thickness of the pixel p. We follow The strategy above calculates the maximum thickness of each pixel on the object's target structure.

Preferably, we can also simultaneously calculate the maximum thickness of each pixel on the object target structure during the process of filling the largest standard pattern. First, the maximum standard pattern is filled with the center of each pixel on the object target structure or the vertices of the pixel, and then for each pixel inside each filled maximum standard pattern, if the pixel currently records the thickness value less than the maximum The thickness of the standard graphic is updated to the thickness of the maximum standard graphic. Then, the thickness value of each pixel finally obtained is the maximum thickness of the pixel obtained by the maximum standard pattern filling method.

7 is a schematic diagram of a layer-by-layer calculation method for two-dimensional thickness and three-dimensional thickness according to an embodiment of the present invention. For a sphere of diameter d3, when performing a layer-by-layer measurement of two-dimensional thickness, first calculate the thickness of the structure using the maximum circular filling method for each layer, and then create a table for each pixel thickness of each layer. The values are counted, that is, at each level, how many thickness values are shared and how many pixels have each thickness value at that level. When measuring the same sphere in three-dimensional thickness layer by layer, we first measure the thickness of each pixel of the whole sphere by the maximum sphere filling method, and then perform layer-by-layer statistics on the thickness of each pixel in each layer, ie in each One level, how many thickness values are recorded and how many pixels have each thickness value at that level. For the same three-dimensional sphere structure, when the two-dimensional thickness is measured layer by layer, since each layer is a circle but the diameters of different layers are not necessarily the same, we may get completely different thickness values at each level. And the number of pixels corresponding to the thickness value; when performing the three-dimensional thickness layer-by-layer measurement, first, the thickness of each pixel of the sphere is first calculated by using the maximum sphere filling method, and each pixel on the sphere has the same thickness value. Then, the thickness of each pixel at each level is counted layer by layer, then each layer has only one thickness value, but the number of pixels having the thickness value may be completely different at different levels. By two-dimensional thickness or three-dimensional thickness layer-by-layer calculation, we not only calculate the thickness of each pixel of the three-dimensional object, but also describe the layer-by-layer distribution of these thicknesses.

FIG. 8 is a schematic diagram of a method for calculating an average thickness of an object according to an embodiment of the present invention. The two-dimensional object to be determined is calculated by the maximum square filling method to calculate the maximum thickness of each pixel on the object. Each of the pixels is represented by a small square, and the object to be measured is composed of shadowed pixels; after being filled by the maximum square method, the number on the pixel represents the calculated maximum thickness of the pixel. Table 2 and Table 3 are the calculation methods of the two-dimensional and three-dimensional average thickness of the target structure in Fig. 8, respectively, where i is the pixel with the maximum thickness i, and S _i is the total number of pixels with the maximum thickness _i in the object. L _i is the thickness of the object corresponding to the pixel length i, [Sigma] S _i for all pixels contained in the object, ΣL _i is the length of the object,

The average thickness of a two- or three-dimensional object.

Fig. 9 is a comparison of results of a two-dimensional image thickness measuring method according to an embodiment of the present invention. The icons Square, Circle, CirSquare, and BoneJ represent the thickness of a standard two-dimensional image measured using BoneJ software, using maximum square fill, maximum circular fill, maximum circular or square fill, respectively. The Circle diagram is the result of various methods for measuring the thickness of a standard circle having a diameter from 1 pixel to 10 pixels. The Square graph is a result of various methods for measuring the thickness of a standard square having a side length of from 1 pixel to 10 pixels. The Rectangle diagram is a result of measuring the thickness of a standard rectangle having a fixed length of 30 pixels and a width of 1 pixel to 10 pixels, respectively. Among them, the maximum circular or square filling method is the most accurate measurement of these three shapes, and there is no error.

Table 2 2D thickness calculation method of target structure

Table 3 Three-dimensional thickness calculation method of target structure

Fig. 10 is a comparison of results of a three-dimensional image thickness measuring method according to an embodiment of the present invention. Icon Cube, Sphere, CubSphere, BoneJ, and Scanco represent the thickness of a standard three-dimensional image measured using maximum cube fill, maximum sphere fill, maximum sphere or cube fill, BoneJ and Scanco software, respectively. The Sphere plot is the result of various methods for measuring the thickness of a standard sphere from 1 pixel to 10 pixels in diameter. The Cube diagram is the result of various methods for measuring the thickness of a standard cube with side lengths from 1 pixel to 10 pixels. The Cuboid diagram is a result of measuring the thickness of a standard rectangular parallelepiped having a fixed length of 30 pixels, a fixed width of 30 pixels, and a height of 1 pixel to 10 pixels, respectively. The Cylinder and Cylinder2 diagrams are the results of various methods for measuring the thickness of a standard cylinder with a fixed height of 30 pixels and a bottom circle diameter of 1 pixel to 10 pixels, respectively. The Cylinder diagram is an average thickness calculation using all the planes of the cylinder. As a result, the Cylinder2 diagram uses various methods to calculate the thickness of each pixel in the cylinder, and selects the interval from the level of the radius of the upper surface of the cylinder to the end of the radius of the bottom surface of the lower surface. The result of the average thickness calculation was performed. The results show that the maximum sphere or cube filling method is the most accurate measurement of these shapes and there is no error.

Figure 11 is a graph showing the effect of osteogenic small molecules on the distribution of trabecular bone thickness from the growth plate along the long axis of the femur in accordance with an embodiment of the present invention. The distal femur samples of the rats were analyzed layer by layer from the plane of the growth plate and the thickness of the two-dimensional and three-dimensional trabecular bones were analyzed layer by layer according to the distance of the growth plates according to different stimulation treatments. Figure 11A shows the two-dimensional thickness (Tb.Th-2D) and three-dimensional thickness (Tb.Th-3D) of trabecular bone in rats with PBS (CTRL), osteogenic small molecule (TR) and PTH (PTH), respectively. influences. Figure 1B shows the trabecular bone two-dimensional thickness (Tb.Th-2D) and three-dimensional thickness (Tb.) of bilateral oophorectomy rats with PBS (CTRL), osteogenic small molecule (TR) and PTH (PTH), respectively. The effect of Th-3D), in which bilateral ovarian blank surgery group (Sham) rats treated with PBS were used as controls. The results show that the two-dimensional thickness and three-dimensional thickness parameters of the trabecular bone have small but very obvious changes under different experimental stimulation conditions, suggesting that the two-dimensional thickness and three-dimensional thickness of the trabecular bone may be a good evaluation of the trabecular bone microstructure. Indicator of change.

In Table 4, after treatment with PBS (CTRL) or osteogenic small molecule (TR), femoral samples were taken and scanned by micro-CT. The area below 0.5 mm to 3.0 mm below the distal femoral growth plate was selected and the parameters of the trabecular bone were calculated and analyzed by multivariate analysis of variance (MANOVA, Figure 12A). The Tb2d.1 to Tb2d.10 parameters represent the volume (number of pixels) of the trabecular bone structure with a thickness of 1 to 10 pixels obtained by the two-dimensional thickness calculation method, and the Tb3d.1 to Tb3d.10 parameter generation. The volume (number of pixels) of the trabecular bone structure portion having a thickness of 1 to 10 pixels, respectively, obtained by the three-dimensional thickness calculation method. In Figure 12, factor analysis or principal component analysis is used to analyze the implicit factors or major components that determine the various parameters of the trabecular bone. The results showed that osteogenic small molecule drugs not only had significant effects on BMC, BV/TV and BMD parameters of trabecular bone, but also significantly affected the two-dimensional thickness parameters of trabecular bone Tb2d.5, Tb2d.6, Tb2d.7, Tb2d.8 and three-dimensional thickness parameters Tb3d.4, Tb3d.5, Tb3d.6 suggest that different trabecular bone two-dimensional or three-dimensional thickness parameters are also sensitive indicators of trabecular bone structure changes. The load diagram of the factor rotation shows that the two main components of the trabecular bone can explain the variance variation of the parameters of the trabecular bone greater than 90% (90% is the principal component extraction threshold of the experimental data of this group), while BV, TV, BMC The BV/TV and BMD parameters are not consistent with the distribution of the thickness parameters of the trabecular bone in the two-dimensional or three-dimensional thickness of the trabecular bone on the load map of the factor rotation, indicating that the various thickness parameters of the trabecular bone provide additional information about the trabecular bone structure, and This additional information cannot be represented by traditional trabecular bone BV, TV, BMC, BV/TV or BMD parameters, suggesting that the thickness parameters of the trabecular bone may be sensitive indicators of trabecular bone structure changes.

Table 4 Effect of osteogenic small molecules on trabecular bone parameters

a: the result of multivariate analysis of variance for image parameters not including Tb3d.1-Tb3d.10;

b: Results of multivariate analysis of variance for image parameters not including Tb2d.1-Tb2d.10.

13 and FIG. 14 are diagrams showing the results of analyzing the different parameters of the trabecular bone including the two-dimensional and three-dimensional thickness by the hierarchical clustering method, respectively, according to an embodiment of the present invention. After the rats were treated with PBS (CTRL) or osteogenic small molecules (TR), the femur samples were taken and scanned by micro-CT. Select the distal femur The parameters of the trabecular bone were calculated in the area of 0.5 mm to 3.0 mm below the long board, and hierarchical analysis was performed on these parameters. The Tb2d.1 to Tb2d.10 parameters represent the volume (number of pixels) of the trabecular bone structure with a thickness of 1 to 10 pixels respectively obtained by the two-dimensional thickness calculation method, and the Tb3d.1 to Tb3d.10 parameters represent three-dimensional The thickness calculation method obtains a volume (number of pixels) of a trabecular bone structure portion having a thickness of 1 to 10 pixels, respectively. The results showed that BV, TV, BMC, BMD, BV/TV and different trabecular bone two-dimensional thickness parameters (Tb2d.1-Tb2d.10, Figure 13) or different trabecular bone three-dimensional thickness parameters (Tb3d.1- Tb3d.10, Figure 14) is clearly grouped into different groups, suggesting that these different groups represent different aspects of the trabecular bone.

On the basis of any of the above embodiments, the present embodiment provides a method for classifying objects according to various image parameters of a target structure, and the method specifically includes: acquiring different objects of different known classifications, and calculating the Knowing each image parameter of the target structure in the scanned image of the classified object and the object to be detected, the image parameter including the thickness parameter of the known classified object and the target structure of the object to be detected; the object according to the known classification and the object target to be detected The individual image parameter values of the structure and the known classification conditions are classified by the one or several statistical methods of discriminant analysis, principal component analysis, factor analysis and logistic regression.

Table 5 is a result of classifying different processing modes of samples by discriminant analysis according to an embodiment of the present invention. For the measured trabecular bone parameters, we set the grouping variables to different small molecule processing methods, and the independent variables are set to BV, TV, BMD or BV, TV, BMD, Tb2d.1-Tb2d.10 or BV, respectively. TV, BMD, Tb3d.1-Tb3d.10, and then discriminant analysis. The results show that the accuracy of grouping using discriminant analysis is only 73.9%, and the correct rate of grouping is 56.5. %. However, when we add the trabecular bone two-dimensional thickness parameter (Tb2d.1-Tb2d.10) or the three-dimensional thickness parameter (Tb3d.1-Tb3d.10) to the argument list of discriminant analysis, the group correct rate and interactive verification analysis The correct rate of grouping is above 90%. For any unclassified new sample, the discriminant analysis statistical method can directly classify the unclassified samples by using the individual trabecular bone parameters of the known classification.

Table 6 is a result of classifying different processing modes of samples by factor analysis according to an embodiment of the present invention. Principal component analysis or factor analysis is widely used in the diagnosis of diseases, credit evaluation, and comprehensive evaluation of economic development. Therefore, principal component analysis or factor analysis is also potential.

Table 5 Discriminant analysis results of classification of different treatment methods of samples

An important method to assist in the diagnosis of osteoporosis. For the measured trabecular bone parameters, we set BV, TV, BMD or BV, TV, BMD, Tb2d.1-Tb2d.10 or BV, TV, BMD, Tb3d.1-Tb3d.10 as variables, and then proceed factor analysis. The results show that the two-dimensional thickness parameters (Tb2d.1-Tb2d.10) or three-dimensional thickness parameters (Tb3d.1) are included in the trabecular bone compared with the BMD or BV, TV, and BMD parameters. The comprehensive evaluation factor of -Tb3d.10) is more accurate than the classification using the factors without these thickness parameters. In the principal component analysis or factor analysis, the expressions of the respective principal components or the respective factors can be obtained by the principal component coefficient matrix or the component score coefficient matrix, respectively. The integrated principal component or the comprehensive factor is calculated by using the proportion of each principal component or factor according to the corresponding variance contribution rate as the weight, and the calculation formula is

Comprehensive

Table 6 Factor analysis results of classification of different treatment methods of samples

Where P _i is the i-th principal component expression, F _i is the i th factor expressions, λ ₁ ... λ _n is the n principal components or factors corresponding to the characteristic root value. Using the reference samples, we can get the expressions of each principal component, each factor, the composite principal component or the composite factor and the values of these parameters in each reference sample. Based on these values, we set the respective thresholds for each principal component, each factor, integrated principal component, and synthesis factor to maximize the discrimination between the small molecule processing group and the control group in the reference sample. For any new sample, the standardization will be standardized. The variables are respectively brought into the expression of each principal component, each factor, integrated principal component or comprehensive factor, and the classification of the new sample can be determined according to the threshold of each parameter.

Table 7 is a logistic regression method for different processing methods of samples according to an embodiment of the present invention. The result of the classification. For the measured trabecular bone parameters, we set the dependent variable to be a different small molecule treatment method, respectively set BV, TV, BMD or BV, TV, BMD, Tb2d.1-Tb2d.10 or BV, TV, BMD, Tb3d.1-Tb3d.10 is a covariate, the variable selection method is set to Forward: LR, and then logistic regression analysis is performed. The results show that the two-dimensional thickness parameters (Tb2d.1-Tb2d.10) or three-dimensional thickness parameters (Tb3d.1-Tb3d) are included in the trabecular bone compared with the BV, TV, and BMD parameters. The logistic regression method of .10) is more accurate than the classification using the thickness parameters. In the logistic regression method, each of the manually selected or variable selected by stepwise regression corresponds to a partial regression coefficient. Therefore, for any new sample, the constant term, the partial regression coefficient and the corresponding variable value are substituted into the regression model, according to the calculation. The result is a classification of the new sample.

In patients with osteoporosis, not only is the bone density reduced, but the microstructural degradation of the trabecular bone is accompanied by a decrease in the thickness of the trabecular bone. However, studies have shown that the bone mineral density values of most fracture patients are not low, and fracture risk assessment tools that combine parameters such as bone density cannot effectively predict the observed fractures for the vast majority of patients. Therefore, the comprehensive detection of bone mineral density and trabecular bone microstructural changes may be important for the diagnosis of osteoporosis and fracture risk assessment.

Table 7 Logistic regression results of classification of different treatment methods of samples

Based on any of the above embodiments, the present embodiment provides a method for assisting in the diagnosis of a disease, which assists in the diagnosis of osteoporosis. First, first measure the parameters of the known and classified vertebral or long bone trabecular bone in the normal population and the patient group, including the thickness parameter; then use one or more of discriminant analysis, principal component analysis, factor analysis, logistic regression The statistical method classifies the subject's osteoporosis state according to various measurement parameters of the vertebral bone or the long bone trabecular bone of the subject.

In Alzheimer's disease, brain weight and brain volume are significantly reduced, the sulcal widens, the cerebral gyrus narrows, and the ventricles enlarge. Magnetic resonance imaging analysis showed that the gray matter, white matter and ventricles in the brain of Alzheimer's patients were significantly different from those of normal controls. Because the gray matter and white matter of the brain are irregularly distributed, it is very similar to trabecular bone. Therefore, the measurement of structural thickness such as gray matter, white matter and ventricle in brain images such as MRI and CT will contribute to the diagnosis of Alzheimer's disease.

Based on any of the above embodiments, the present embodiment provides a method for assisting in the diagnosis of diseases, which assists in the diagnosis of Alzheimer's disease. First, the parameters of the gray matter, white matter and ventricles in the normal population and the patient group of the known classification, including the thickness parameter, are first measured; then one of discriminant analysis, principal component analysis, factor analysis, logistic regression or Several statistical methods classify the subject's Alzheimer's state according to the measured parameters of the subject's gray matter, white matter and ventricles.

It should be understood that portions not specifically described in this specification are prior art.

It should be understood that the above description of the preferred embodiments is not to be construed as limiting the scope of the present invention. In the case of the scope of the invention, it is intended that modifications and variations may be made without departing from the scope of the invention.

Claims

A method for detecting changes in image structure, comprising the steps of:

Step 1: acquiring a region of interest of each image of the object to be detected, and determining respective image parameter values inside the region of interest of each image of the object to be detected, the image parameter value including the total volume of the region of interest, the target structure volume One or more image parameter values of the target structure material content, the thickness of the target structure, the target structure volume fraction, and the target structure density;

Step 2: statistically analyzing, according to each image parameter value of each region of the image of the object to be detected, a difference saliency of each image parameter value of the region of interest of the object to be detected in different classifications or for the object to be detected sort.
The method for detecting a change in an image structure according to claim 1, wherein the step of acquiring the region of interest of each of the three-dimensional object to be detected in step 1 comprises:

Step A.1: measuring each image parameter value of each three-dimensional object in different groupings in its respective two-dimensional level region of interest;

Step A.2: Selecting one or more layers of two-dimensional layers from each of the three-dimensional objects, and statistically analyzing the significance test of each image parameter value of the selected two-dimensional level region of interest between different groups. value;

Step A.3: statistically analyzing the significance value of each image parameter value of the three-dimensional object in all possible two-dimensional regions of interest between different groups according to the method described in step A.2;

Step A.4: Select the minimum p value in step A.3 or the p value less than the preset threshold, and determine the region of interest of the corresponding three-dimensional object according to the selected p value.
The method for detecting a change in image structure according to claim 1, wherein the thickness of the target structure is determined in step 1, by a maximum square filling method, a maximum circular filling method, a maximum circular or square filling method, and a maximum cube. One or more of the filling method, the maximum sphere filling method, the maximum sphere or the cube filling method are calculated.
The method of detecting a change in image structure according to claim 3, wherein when the thickness of each pixel in the target structure is measured by a maximum square filling method, a maximum circular filling method, a maximum cube filling method, or a maximum sphere filling method The center of the filled largest shape containing the pixel to be determined and completely contained in the target structure is located at the center of the pixel on the target structure or at the apex of the pixel.
A method of detecting a change in image structure according to claim 3, wherein when the thickness of the target structure is measured by a maximum circular or square filling method, said maximum square is used for each pixel in said target structure The maximum value of the thickness of the pixel obtained by the filling method and the maximum circular filling method is taken as the thickness of the pixel.
A method of detecting a change in image structure according to claim 3, wherein when the thickness of the target structure is measured by a maximum sphere or cube filling method, said maximum sphere is filled for each pixel in said target structure The method and the maximum value of the thickness of the pixel obtained by the maximum cube filling method are taken as the thickness of the pixel.
The method for detecting a change in image structure according to claim 1, wherein the statistically analyzing the difference significance of the target structure in different classifications in the step 2 includes: using t test, one-way analysis of variance, and linear regression. Analysis, non-linear regression analysis, nonlinear exponential decay regression analysis, logistic regression, repeated measurement analysis of variance, and linear mixed-effects model repeated measurement analysis of one or several statistical methods for each image of the target structure of the object to be detected The parameters were analyzed for differences in significance under different classification conditions.
The method for detecting a change in an image structure according to claim 1, wherein the classifying the object to be detected in the step 2 comprises: acquiring different objects of a known classification, and calculating the known classification. Each image parameter of the target structure in the scanned image of the object and the object to be detected, the image parameter including the thickness of the known classified object and the target structure of the object to be detected; each of the object according to the known classification and the target structure of the object to be detected The image parameter values and the known classification conditions are used to classify the objects to be detected by one or several statistical methods of discriminant analysis, principal component analysis, factor analysis, and logistic regression.
The method for detecting a change in image structure according to claim 8, wherein the classifying the object to be detected is to classify the severity of osteoporosis of the bone to be measured.
The method for detecting a change in image structure according to claim 8, wherein the classifying the object to be detected is classifying the severity of Alzheimer's disease of the individual to be measured.