CN110555853B - Method and device for segmentation algorithm evaluation based on anatomical priors - Google Patents
Method and device for segmentation algorithm evaluation based on anatomical priors Download PDFInfo
- Publication number
- CN110555853B CN110555853B CN201910728421.9A CN201910728421A CN110555853B CN 110555853 B CN110555853 B CN 110555853B CN 201910728421 A CN201910728421 A CN 201910728421A CN 110555853 B CN110555853 B CN 110555853B
- Authority
- CN
- China
- Prior art keywords
- segmentation
- shape
- value
- obtaining
- prior
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30061—Lung
Abstract
The invention provides a method and a device for segmentation algorithm evaluation based on anatomical prior, wherein the method comprises the following steps: carrying out shape processing on the relative position information of the central point coordinates of each segmented target area according to an active shape model algorithm and a Purchase analysis method to obtain a processed shape; calculating the average value of the processed shape, the eigenvalue of the covariance matrix and the eigenvector value of the covariance matrix according to the data of the training sample set; obtaining a shape coefficient value according to the average value of the processed shape and the characteristic vector value of the covariance matrix; obtaining a prior with stable relative position of the total segmentation target area according to the shape coefficient value and the characteristic value of the covariance matrix; obtaining the unique prior of each segmentation target region according to the segmentation deviation value of each segmentation target region; and obtaining a training model according to the stable prior of the relative position of the total segmentation target region and the unique prior of each segmentation target region, and evaluating the current segmentation algorithm.
Description
Technical Field
The invention relates to the field of medical images, in particular to a segmentation algorithm evaluation method and device based on anatomy prior.
Background
Medical image analysis methods based on deep learning models are increasingly gaining attention. In order to achieve good analysis effect, a large amount of data needs to be labeled for model training and testing. Since high-quality medical data labeling requires a doctor to have a rich experience, under the condition of limited labeling resources, how to select the data with the most information amount for labeling by evaluating the effect of the existing model is important.
The existing segmentation algorithm evaluation methods are mainly divided into two categories. One is to directly calculate an evaluation index value according to the difference between the segmentation result and the label under the condition of the label, but the method is obviously not suitable for the data which is not labeled. The other method mainly extracts various image features such as shape features and texture features according to segmentation results, and trains a nonlinear model by using the features to approximate and estimate evaluation index values. Two problems exist in this way, namely, the training time is long, so that the model is updated in an iterative manner slowly; secondly, a large number of features need to be designed manually, but the prior anatomical information of the task itself is not fully utilized.
Therefore, how to find a mode suitable for model effect evaluation and data selection based on the anatomical structure becomes a problem which needs to be solved urgently at present.
Disclosure of Invention
In order to fully utilize the anatomical prior information of a task and find data with large information quantity to evaluate the segmentation effect of a model, the invention provides a segmentation algorithm evaluation method and device based on anatomical prior.
In a first aspect, the present invention provides a method for segmentation algorithm evaluation based on anatomical priors, the method comprising:
carrying out shape processing on the relative position information of the central point coordinates of each segmented target area according to an active shape model algorithm and a Purchase analysis method to obtain a processed shape;
calculating the average value of the processed shape, the eigenvalue of the covariance matrix and the eigenvector value of the covariance matrix according to the data of the training sample set;
obtaining a shape coefficient value according to the average value of the processed shape and the characteristic vector value of the covariance matrix;
obtaining a prior with stable relative position of the total segmentation target area according to the shape coefficient value and the characteristic value of the covariance matrix;
obtaining the unique prior of each segmentation target region according to the segmentation deviation value of each segmentation target region;
and obtaining a training model according to the stable prior of the relative position of the total segmentation target region and the unique prior of each segmentation target region, and evaluating the current segmentation algorithm.
Further, the segmentation target region is a lung lobe segmentation region.
Further, performing shape processing on the relative position information of the center point coordinates of each segmented target region according to an active shape model algorithm and a Fourier analysis method, and obtaining the processed shape comprises the following steps:
acquiring training sample set data and relative position information of coordinates of central points of all lung lobes;
performing shape modeling on the relative position information by using an active shape model algorithm;
and carrying out normalization processing on the obtained shape by utilizing a Pushing analysis method to obtain the processed shape.
Further, a training model is obtained according to the stable prior of the relative position of the total segmentation target region and the unique prior of each segmentation target region, and the current segmentation algorithm is evaluated by the following steps:
obtaining data added into the training model according to the stable prior of the relative position of the total lung lobes and the unique prior of each lung lobe;
and inputting the data of the test sample set into the training model added with the data, and evaluating the current segmentation algorithm.
Further, obtaining a unique prior of each segmented target region according to the segmentation bias value of each segmented target region includes:
and obtaining the unique prior of each lung lobe according to the mean value of the number of the segmentation deviations of each lung lobe and the standard deviation of the number of the segmentation deviations of each lung lobe.
Further, inputting the test sample set data into the training model after adding the data, and evaluating the current segmentation algorithm comprises:
smoothing the training model added with the data according to a smoothing edge loss function;
inputting the data of the test sample set into the training model after the smoothing treatment;
and evaluating the current segmentation algorithm by using the obtained training result.
Further, according to the average value of the processed shape and the characteristic vector value of the covariance matrix, a calculation formula for obtaining a shape coefficient value is as follows:
wherein the content of the first and second substances,
represents an average value of the processed shape, and x represents the processed shape; v. ofiAn ith eigenvector representing a covariance matrix; alpha is alphaiRepresenting a shape factor value; i represents the number of lung lobes, and i is more than or equal to 1 and less than or equal to k.
Further, according to the shape coefficient value and the eigenvalue of the covariance matrix, the prior calculation formula for obtaining the relative position stability of the total segmentation target area is as follows:
wherein alpha isiRepresenting a shape factor value; sigmaiAn ith eigenvalue representing a covariance matrix; i represents the number of lung lobes, and i is more than or equal to 1 and less than or equal to k.
Further, according to the mean value of the number of the segmentation deviations of each lung lobe and the standard deviation of the number of the segmentation deviations of each lung lobe, a unique prior calculation formula of each lung lobe is obtained as follows:
wherein the content of the first and second substances,
cirepresenting the number of components of a single lung lobe;
representing the mean value of the number of the segmentation deviations of each lung lobe; sigmanThe number standard deviation of each division deviation is shown.
In a second aspect, the present invention provides an apparatus for segmentation algorithm evaluation based on anatomical priors, the apparatus comprising:
the processing shape module is used for carrying out shape processing on the relative position information of the central point coordinates of each segmented target area according to an active shape model algorithm and a Prussian analysis method to obtain a processed shape;
the calculation module is used for calculating the average value of the processed shape, the eigenvalue of the covariance matrix and the eigenvector value of the covariance matrix according to the data of the training sample set;
a shape coefficient value obtaining module for obtaining a shape coefficient value according to the processed average value of the shape and the characteristic vector value of the covariance matrix;
the prior obtaining module is used for obtaining the prior with stable relative position of the total segmentation target area according to the shape coefficient value and the characteristic value of the covariance matrix;
the unique priori obtaining module of each segmented target region is used for obtaining the unique priori of each segmented target region according to the segmentation deviation value of each segmented target region;
and the evaluation module is used for obtaining a training model according to the stable prior of the relative position of the total segmentation target area and the unique prior of each segmentation target area and evaluating the current segmentation algorithm.
According to the invention, the segmentation evaluation based on anatomical priors is completed aiming at the segmentation target areas with fixed position relation, and the data with large information amount is selected through the two indexes of the prior with stable relative position of the total segmentation target area and the unique prior of each segmentation target area, so as to evaluate the model effect and further obtain the effectiveness of the current segmentation algorithm.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the embodiments or the prior art descriptions will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a diagram illustrating the segmentation of lung lobes according to the prior art; wherein, 1(a) in fig. 1 represents an original image, 1(b) in fig. 1 represents an artificial segmentation lung lobe label, and 1(c) in fig. 1 represents a lung lobe segmentation prediction result;
FIG. 2 is a flow chart of a method for segmentation algorithm evaluation based on anatomical priors provided by an embodiment of the present invention;
fig. 3 is a block diagram of an apparatus for segmentation algorithm evaluation based on anatomical priors according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The existing segmentation algorithm evaluation method does not fully utilize the anatomical prior information of the task, so that the training time is long, the model iteration is updated slowly, a large amount of information is labeled manually, and a lung lobe segmentation result violating the anatomical prior can occur. As shown in fig. 1, 1(a) in fig. 1 represents an original image, 1(b) in fig. 1 represents an artificial labeling diagram, and 1(c) in fig. 1 represents a prediction result. It can be seen that the segmentation of the lung lobes into multiple parts in the prediction result of 1(c) in fig. 1 is in severe violation of the anatomy-prior lung lobe segmentation. In order to solve the above problem, the present invention provides a method and apparatus for segmentation algorithm evaluation based on anatomical priors, as shown in fig. 2, the method comprising:
step S201, according to an active shape model algorithm and a Prussian analysis method, shape processing is carried out on the relative position information of the central point coordinates of each segmentation target area, and the processed shape is obtained.
Specifically, the method and the device are used for the segmentation algorithm evaluation based on the anatomical prior, wherein the segmentation algorithm evaluation is fixed in the position relation such as human body segmentation, brain region segmentation and lung lobe segmentation. Here, taking the lung lobe segmentation as an example, the details are: a typical human lung is divided into 5 lobes by 3 interfilamentary divisions. Wherein the right lung typically comprises 3 lobes (upper right, middle right, lower right) and the left lung typically comprises two lobes (upper left and lower left), and the relative positions between the lobes are stable, although the lobes are each shaped differently. And carrying out shape processing on the relative position information of the coordinates of the central point of each lung lobe according to an active shape model algorithm and a Purchase analysis method to obtain a processed shape. The active shape model algorithm is a probability model which expresses the shape of a certain object into a flexible curve, the curve can be appropriately deformed for matching with a specific example of the object, the relative position of the coordinates of the central points of 5 lung lobes is used as one shape in the active shape model algorithm, and the shape obtained by the active shape model algorithm is processed according to a Pouler analysis method.
Step S202, calculating the average value of the processed shape, the eigenvalue of the covariance matrix and the eigenvector value of the covariance matrix according to the training sample set data.
Specifically, the average value of the processed shape, the eigenvalue of the covariance matrix, and the eigenvector value of the covariance matrix are calculated according to the collected training sample set data. It can be understood that, taking the training sample set data as a matrix, obtaining the eigenvalue of the covariance matrix and the eigenvector value of the covariance matrix are common calculation methods.
Step S203, obtaining a shape coefficient value according to the processed shape average value and the characteristic vector value of the covariance matrix.
Specifically, the shape coefficient value is obtained from the obtained average value of the processed shape and the characteristic vector value of the covariance matrix.
And step S204, obtaining the prior with stable relative position of the total segmentation target area according to the shape coefficient value and the characteristic value of the covariance matrix.
Specifically, here, a priori that the relative position of the total lung lobes is stable can be obtained according to the shape coefficient value and the eigenvalue of the covariance matrix. Here, the total number of lobes is 5.
Step S205, obtaining the unique prior of each segmentation target area according to the segmentation deviation value of each segmentation target area.
Specifically, since each lung lobe is unique, it is not reasonable if 0 or more components are segmented in the segmentation result, and thus, a unique prior for each lung lobe is obtained according to the segmentation deviation value of each lung lobe.
And S206, obtaining a training model according to the stable prior of the relative position of the total segmentation target area and the unique prior of each segmentation target area, and evaluating the current segmentation algorithm.
Specifically, in the embodiment of the invention, lung lobe segmentation is used as a background, and data which is large in information amount and relatively difficult to train a model are selected according to two indexes, namely a stable prior of the relative position of total lung lobes and a unique prior of each lung lobe, and added into the training model to obtain the training model, so that the effectiveness of the current segmentation algorithm is evaluated.
It should be noted that the embodiment of the present invention does not specifically limit the type to be segmented, and as long as the position relationship of the type to be segmented is fixed, the embodiment of the present invention is applicable to the segmentation evaluation step based on anatomical priors, which is proposed in the embodiment of the present invention.
The invention finishes the segmentation evaluation based on anatomy prior aiming at the segmentation target areas with fixed position relation, and selects data with large information amount through two indexes of prior with stable relative position of the total segmentation target area and unique prior of each segmentation target area for evaluating the model effect, thereby obtaining the effectiveness of the current segmentation algorithm.
Based on the content of the above embodiments, as an alternative embodiment: according to an active shape model algorithm and a Purchase analysis method, carrying out shape processing on the relative position information of the central point coordinates of each segmented target area, and obtaining the processed shape comprises the following steps:
acquiring training sample set data and relative position information of coordinates of central points of all lung lobes;
performing shape modeling on the relative position information by using an active shape model algorithm;
and carrying out normalization processing on the obtained shape by utilizing a Pushing analysis method to obtain the processed shape.
Specifically, the embodiment of the present invention also takes lung lobe segmentation as an example for explanation: here, an active shape model represents the shape of an object of a certain class as a probabilistic model of a flexible curve that can be appropriately deformed for matching with a specific instance of the object of the class. "Internal energy" is used to quantify the reasonable degree of such deformation, and its value is inversely proportional to the reasonable degree of deformation. The active shape model connects the relative position information of the coordinates of the central points of the lung lobes into a shape, and normalization processing is carried out on the shape through scaling, rotation and translation by utilizing a Poincare analysis method to obtain the normalized shape.
Based on the content of the above embodiments, as an alternative embodiment: obtaining a training model according to the stable prior of the relative position of the total segmentation target region and the unique prior of each segmentation target region, wherein the current segmentation algorithm comprises the following steps:
obtaining data added into the training model according to the stable prior of the relative position of the total lung lobes and the unique prior of each lung lobe;
and inputting the data of the test sample set into the training model added with the data, and evaluating the current segmentation algorithm.
Specifically, data obtained by stable prior of the relative position of the total lung lobes and unique prior of each lung lobe is used as data which is large in information amount and difficult to train relative to a training model, and then the data of the test sample set is input into the training model added with the selected data to obtain a training result, and then the current segmentation algorithm is evaluated.
Based on the content of the above embodiments, as an alternative embodiment: obtaining the only prior of each segmented target region according to the segmentation deviation value of each segmented target region comprises:
and obtaining the unique prior of each lung lobe according to the mean value of the number of the segmentation deviations of each lung lobe and the standard deviation of the number of the segmentation deviations of each lung lobe.
Specifically, when the unique prior of each lung lobe is obtained, the specific parameter values used are the mean value of the number of segmentation deviations of each lung lobe and the standard deviation of the number of segmentation deviations.
Based on the content of the above embodiments, as an alternative embodiment: inputting the data of the test sample set into a training model added with the data, and evaluating the current segmentation algorithm comprises the following steps:
smoothing the training model added with the data according to a smoothing edge loss function;
inputting the data of the test sample set into the training model after the smoothing treatment;
and evaluating the current segmentation algorithm by using the obtained training result.
Specifically, the training model with the data added is smoothed according to a smoothing edge loss function, that is, in the training model forming process, the loss data is correspondingly processed, so that the obtained training model has higher segmentation accuracy. Inputting the test sample set data into the training model after the smoothing treatment, and evaluating the effectiveness of the current segmentation algorithm according to the training result obtained by the training model after the smoothing treatment.
According to the non-zero surrounding rule of computer graphics, the number of the image edges through which a straight line passes is in direct proportion to the number of convex components. By observing the probability chart output by the model, the edge in the lung lobe segmentation result is obtained due to huge jump in probability. Thus, it is concluded that reducing the number of unreasonable probability hops can eliminate unreasonable components. Therefore, in the embodiment of the present invention, with lung lobe segmentation as a background, a smooth edge loss function is defined as follows:
wherein the content of the first and second substances,
it should be noted that X and Y are respectively a predicted probability map and a unique hot code label of each lung lobe, and the unique hot code is generally used when labeling a data set in supervised learning. That is, X represents training data, and Y represents annotation data selected from the training data; c is a class subscript; c represents 1-5 lobe classes, D represents depth, W represents width, and H represents height; i, j, k are the voxel indices md(XcI, j, k) is a scalar edge loss calculated from the gradient of d direction, where d has three directions, x, y, z, respectively; w is ad(YcI, j, k) is a weight mask for eliminating edge loss of the region where the labeling class changes, λ is a "sudden change" probability value that may generate class jump in the segmentation result, and is a value preset according to the actual situation, and preferably, when λ is 0.05 or λ is 0.10, the model effect is the best.
Based on the content of the above embodiments, as an alternative embodiment: according to the processed average value of the shape and the characteristic vector value of the covariance matrix, a calculation formula for obtaining a shape coefficient value is as follows:
wherein the content of the first and second substances,
represents an average value of the processed shape, and x represents the processed shape; v. ofiAn ith eigenvector representing a covariance matrix; alpha (alpha) ("alpha")iRepresenting a shape factor value; i represents the number of the segmentation target areas, and i is more than or equal to 1 and less than or equal to k.
Based on the content of the above embodiments, as an alternative embodiment: according to the shape coefficient value and the eigenvalue of the covariance matrix, the prior calculation formula for obtaining the relative position stability of the total lung lobes is as follows:
wherein alpha isiRepresenting a shape factor value; sigmaiAn ith eigenvalue representing a covariance matrix; i represents the number of lung lobes, and i is more than or equal to 1 and less than or equal to k.
Based on the content of the above embodiments, as an alternative embodiment: according to the mean value of the segmentation deviation number of each lung lobe and the standard deviation of the segmentation deviation number of each lung lobe, a unique priori calculation formula of each lung lobe is obtained as follows:
wherein the content of the first and second substances,
cirepresenting the number of components of a single lung lobe;
indicating the number of segmentation deviations per lung lobeMean value; sigmanThe number standard deviation of each division deviation is shown.
Based on the content of the above embodiments, as an alternative embodiment: structural introduction about a lung lobe segmentation model: the lung lobe segmentation model is a 3-dimensional complete convolution neural network, and converts an original lung CT image into a pixel-by-pixel lung lobe class prediction. The following takes a typical 3-dimensional encoding-decoding network as an example: the coding network consists of 4 stages, each of which consists of 2 convolutional layers of step size 1 and 1 pooling layer of step size 2. The decoding network also consists of 4 stages, each consisting of 2 convolutional layers of step size 1 and 1 upsampling layer. The output of each stage of the coding network is directly input into the corresponding stage of the decoding network through a hopping connection. All convolutional layers use a 3 x 3 convolutional kernel followed by a normalization layer and leaky linear rectification unit. To increase the location information of the network, the convolutional layer of the last stage of the decoding network uses a coordinate convolutional layer instead of a normal convolutional layer.
It should be noted that the embodiments of the present invention are not limited to the network structure, and any convolutional neural network suitable for performing the segmentation task is within the scope of the embodiments of the present invention.
Based on the content of the above embodiments, as an alternative embodiment: in the embodiment of the invention, another loss function is defined and used for modeling the training model. The method specifically comprises the following steps: and in the training stage, a lung CT image and corresponding lung lobe mark information are input. And obtaining lung lobe class prediction information by the voxels inside and outside each lung lobe through a segmentation model, and then training by using a loss function. The loss function herein includes smooth edge loss and multi-class Dice loss, where Dice loss is defined as follows:
wherein the content of the first and second substances,
and
respectively is a predicted value of a voxel (i, j, k) and a one-hot coding labeling value of each lung lobe, c is a category subscript, and c is more than or equal to 0 and less than or equal to 5; i, j, k are the voxel indices,. epsilon.takes 0.00001, preventing the denominator from being 0.
And in the testing stage, a lung CT image is input, and a predicted lung lobe class probability map can be obtained through a trained lung lobe segmentation model. Finally, the class of each voxel is the lobe of the lung that takes the highest probability.
According to another aspect of the present invention, the embodiment of the present invention further provides an apparatus for segmentation algorithm evaluation based on anatomical priors, and referring to fig. 3, fig. 3 is a block diagram of the apparatus for segmentation algorithm evaluation based on anatomical priors provided by the embodiment of the present invention. The device is used for completing segmentation algorithm evaluation based on anatomical priors in the previous embodiments. Therefore, the description and definition in the method for segmentation algorithm evaluation based on anatomical priors in the foregoing embodiments can be used for understanding the execution modules in the embodiments of the present invention.
A processing shape module 301, configured to perform shape processing on the relative position information of the center point coordinates of each segmented target region according to an active shape model algorithm and a pockels analysis method to obtain a processed shape;
a calculating module 302, configured to calculate, according to the training sample set data, an average value of the processed shape, an eigenvalue of the covariance matrix, and an eigenvector value of the covariance matrix;
a shape coefficient value obtaining module 303, configured to obtain a shape coefficient value according to the average value of the processed shape and the feature vector value of the covariance matrix;
a prior obtaining module 304 for obtaining the relative position stability of the total segmented target area according to the shape coefficient value and the eigenvalue of the covariance matrix;
a unique prior obtaining module 305 for each segmented target region, configured to obtain a unique prior of each segmented target region according to the segmentation deviation value of each segmented target region;
and the evaluation module 306 is used for obtaining a training model according to the stable prior of the relative position of the total segmentation target region and the unique prior of each segmentation target region, and evaluating the current segmentation algorithm.
According to the invention, the segmentation evaluation based on anatomical priors is completed aiming at the segmentation target areas with fixed position relation, and the data with large information amount is selected through the two indexes of the prior with stable relative position of the total segmentation target area and the unique prior of each segmentation target area, so as to evaluate the model effect and further obtain the effectiveness of the current segmentation algorithm. The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and the parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods of the various embodiments or some parts of the embodiments.
Finally, the principle and the implementation of the present invention are explained by applying the specific embodiments in the present invention, and the above description of the embodiments is only used to help understanding the method and the core idea of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.
Claims (5)
1. A method for segmentation algorithm evaluation based on anatomical priors, the method comprising:
carrying out shape processing on the relative position information of the central point coordinates of each segmented target area according to an active shape model algorithm and a Purchase analysis method to obtain a processed shape;
calculating the average value of the processed shape, the eigenvalue of the covariance matrix and the eigenvector value of the covariance matrix according to the data of the training sample set;
obtaining a shape coefficient value according to the average value of the processed shape and the characteristic vector value of the covariance matrix;
obtaining a prior with stable relative position of a total segmentation target area according to the shape coefficient value and the characteristic value of the covariance matrix; wherein, the calculation formula is:
wherein alpha isiRepresenting a shape factor value; sigmaiAn ith eigenvalue representing a covariance matrix; i represents the number of lung lobes, i is more than or equal to 1 and less than or equal to k;
obtaining the unique prior of each segmentation target region according to the segmentation deviation value of each segmentation target region; the method comprises the following steps: obtaining the unique prior of each lung lobe according to the mean value of the number of the segmentation deviations of each lung lobe and the standard deviation of the number of the segmentation deviations of each lung lobe, wherein the calculation formula is as follows:
wherein the content of the first and second substances,
cirepresenting the number of components of a single lung lobe;
representing the mean value of the number of the segmentation deviations of each lung lobe; sigmanExpressing the number standard deviation of each segmentation deviation;
obtaining a training model according to the stable prior of the relative position of the total segmentation target area and the unique prior of each segmentation target area, and evaluating the current segmentation algorithm; the segmentation target area is a lung lobe segmentation area; obtaining a training model according to the stable prior of the relative position of the total segmentation target region and the unique prior of each segmentation target region, wherein the evaluation of the current segmentation algorithm comprises the following steps: obtaining data added into the training model according to the stable prior of the relative position of the total lung lobes and the unique prior of each lung lobe; and inputting the data of the test sample set into the training model added with the data, and evaluating the current segmentation algorithm.
2. The method according to claim 1, wherein the performing shape processing on the relative position information of each segmented target region according to an active shape model algorithm and a pockels analysis method to obtain a processed shape comprises:
acquiring training sample set data and relative position information of coordinates of central points of lung lobes;
performing shape modeling on the relative position information by using an active shape model algorithm;
and carrying out normalization processing on the obtained shape by utilizing a Pushing analysis method to obtain the processed shape.
3. The method of claim 1, wherein the inputting the test sample set data into the trained model after adding the data, and the evaluating the current segmentation algorithm comprises:
smoothing the training model added with the data according to a smoothing edge loss function;
inputting the data of the test sample set into the training model after smoothing treatment;
and evaluating the current segmentation algorithm by using the obtained training result.
4. The method according to claim 1, wherein the calculation formula for obtaining the shape coefficient value according to the average value of the processed shape and the characteristic vector value of the covariance matrix is as follows:
wherein, the first and the second end of the pipe are connected with each other,
represents the average value of the processed shape, and x represents the numerical value of the processed shape coefficient; v. ofiAn ith eigenvector representing a covariance matrix; alpha is alphaiRepresenting a shape factor value; i represents the number of lung lobes, and i is more than or equal to 1 and less than or equal to k.
5. An apparatus for segmentation algorithm evaluation based on anatomical priors, the apparatus comprising:
the processing shape module is used for carrying out shape processing on the relative position information of the central point coordinates of each segmentation target area according to an active shape model algorithm and a Purchase analysis method to obtain a processed shape;
the calculation module is used for calculating the average value of the processed shape, the eigenvalue of the covariance matrix and the eigenvector value of the covariance matrix according to the training sample set data;
a shape coefficient value obtaining module, configured to obtain a shape coefficient value according to the average value of the processed shape and the feature vector value of the covariance matrix;
and the prior obtaining module is used for obtaining the prior with stable relative position of the total segmentation target area according to the shape coefficient value and the characteristic value of the covariance matrix, wherein the calculation formula is as follows:
wherein alpha isiRepresenting a shape factor value; sigmaiAn ith eigenvalue representing a covariance matrix; i represents the number of lung lobes, i is more than or equal to 1 and less than or equal to k;
the unique priori obtaining module of each segmented target region is used for obtaining the unique priori of each segmented target region according to the segmentation deviation value of each segmented target region, and comprises the following steps: obtaining the unique prior of each lung lobe according to the mean value of the segmentation deviation number of each lung lobe and the standard deviation of the segmentation deviation number of each lung lobe, wherein the calculation formula is as follows:
wherein the content of the first and second substances,
cirepresenting the number of components of a single lung lobe;
representing the mean value of the number of the segmentation deviations of each lung lobe; sigmanExpressing the standard deviation of the number of each segmentation deviation;
and the evaluation module is used for obtaining a training model according to the stable prior of the relative position of the total segmentation target area and the unique prior of each segmentation target area and evaluating the current segmentation algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910728421.9A CN110555853B (en) | 2019-08-07 | 2019-08-07 | Method and device for segmentation algorithm evaluation based on anatomical priors |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910728421.9A CN110555853B (en) | 2019-08-07 | 2019-08-07 | Method and device for segmentation algorithm evaluation based on anatomical priors |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110555853A CN110555853A (en) | 2019-12-10 |
| CN110555853B true CN110555853B (en) | 2022-07-19 |
Family
ID=68737106
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910728421.9A Active CN110555853B (en) | 2019-08-07 | 2019-08-07 | Method and device for segmentation algorithm evaluation based on anatomical priors |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110555853B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111738349B (en) * | 2020-06-29 | 2023-05-02 | 重庆紫光华山智安科技有限公司 | Detection effect evaluation method and device of target detection algorithm, storage medium and equipment |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103310449A (en) * | 2013-06-13 | 2013-09-18 | 沈阳航空航天大学 | Lung segmentation method based on improved shape model |
| CN107146222A (en) * | 2017-04-21 | 2017-09-08 | 华中师范大学 | Medical Image Compression algorithm based on human anatomic structure similitude |
| CN107369160A (en) * | 2017-06-28 | 2017-11-21 | 苏州比格威医疗科技有限公司 | A kind of OCT image median nexus film new vessels partitioning algorithm |
| CN107492105A (en) * | 2017-08-11 | 2017-12-19 | 深圳市旭东数字医学影像技术有限公司 | A kind of variation dividing method based on more statistical informations |
| CN109003278A (en) * | 2018-03-26 | 2018-12-14 | 天津工业大学 | A kind of improved CT image aorta segmentation method based on active shape model |
| CN109242864A (en) * | 2018-09-18 | 2019-01-18 | 电子科技大学 | Image segmentation result quality evaluating method based on multiple-limb network |
| CN109272510A (en) * | 2018-07-24 | 2019-01-25 | 清华大学 | The dividing method of tubular structure in a kind of 3 d medical images |
| CN109598727A (en) * | 2018-11-28 | 2019-04-09 | 北京工业大学 | A kind of CT image pulmonary parenchyma three-dimensional semantic segmentation method based on deep neural network |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10628659B2 (en) * | 2017-11-27 | 2020-04-21 | International Business Machines Corporation | Intelligent tumor tracking system |
| US10607114B2 (en) * | 2018-01-16 | 2020-03-31 | Siemens Healthcare Gmbh | Trained generative network for lung segmentation in medical imaging |
-
2019
- 2019-08-07 CN CN201910728421.9A patent/CN110555853B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103310449A (en) * | 2013-06-13 | 2013-09-18 | 沈阳航空航天大学 | Lung segmentation method based on improved shape model |
| CN107146222A (en) * | 2017-04-21 | 2017-09-08 | 华中师范大学 | Medical Image Compression algorithm based on human anatomic structure similitude |
| CN107369160A (en) * | 2017-06-28 | 2017-11-21 | 苏州比格威医疗科技有限公司 | A kind of OCT image median nexus film new vessels partitioning algorithm |
| CN107492105A (en) * | 2017-08-11 | 2017-12-19 | 深圳市旭东数字医学影像技术有限公司 | A kind of variation dividing method based on more statistical informations |
| CN109003278A (en) * | 2018-03-26 | 2018-12-14 | 天津工业大学 | A kind of improved CT image aorta segmentation method based on active shape model |
| CN109272510A (en) * | 2018-07-24 | 2019-01-25 | 清华大学 | The dividing method of tubular structure in a kind of 3 d medical images |
| CN109242864A (en) * | 2018-09-18 | 2019-01-18 | 电子科技大学 | Image segmentation result quality evaluating method based on multiple-limb network |
| CN109598727A (en) * | 2018-11-28 | 2019-04-09 | 北京工业大学 | A kind of CT image pulmonary parenchyma three-dimensional semantic segmentation method based on deep neural network |
Non-Patent Citations (2)
| Title |
|---|
| Suggestive annotation: A deep active;Lin Yang et al;《 MICCAI》;20171231;全文 * |
| 基于序列图像分析的医学CT图像分割算法研究;揭萍;《中国优秀博硕士学位论文全文数据库(硕士)》;20190115(第12期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110555853A (en) | 2019-12-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20230104173A1 (en) | Method and system for determining blood vessel information in an image | |
| CN109003267B (en) | Computer-implemented method and system for automatically detecting target object from 3D image | |
| CN109102511B (en) | Cerebrovascular segmentation method, system and electronic equipment | |
| Li et al. | Automated measurement network for accurate segmentation and parameter modification in fetal head ultrasound images | |
| CN111640120A (en) | Pancreas CT automatic segmentation method based on significance dense connection expansion convolution network | |
| CN111860528B (en) | Image segmentation model based on improved U-Net network and training method | |
| Wang et al. | Uncertainty-guided efficient interactive refinement of fetal brain segmentation from stacks of MRI slices | |
| CN113095333B (en) | Unsupervised feature point detection method and unsupervised feature point detection device | |
| CN113822851A (en) | Image segmentation method, device, equipment and storage medium | |
| Zhang et al. | Lesion synthesis to improve intracranial hemorrhage detection and classification for CT images | |
| CN112183541A (en) | Contour extraction method and device, electronic equipment and storage medium | |
| CN115409879A (en) | Data processing method and device for image registration, storage medium and electronic equipment | |
| CN113763371A (en) | Pathological image cell nucleus segmentation method and device | |
| Tran et al. | Fully convolutional neural network with attention gate and fuzzy active contour model for skin lesion segmentation | |
| CN115496720A (en) | Gastrointestinal cancer pathological image segmentation method based on ViT mechanism model and related equipment | |
| CN110555853B (en) | Method and device for segmentation algorithm evaluation based on anatomical priors | |
| CN113764101A (en) | CNN-based breast cancer neoadjuvant chemotherapy multi-modal ultrasonic diagnosis system | |
| CN111652876B (en) | Method for detecting three-dimensional basin bottom ultrasonic image | |
| CN112381822A (en) | Method for processing images of focal zones of the lungs and related product | |
| CN115439478B (en) | Pulmonary lobe perfusion intensity assessment method, system, equipment and medium based on pulmonary perfusion | |
| CN115100731B (en) | Quality evaluation model training method and device, electronic equipment and storage medium | |
| CN113177938B (en) | Method and device for segmenting brain glioma based on circular convolution kernel and related components | |
| Yang et al. | A new approach to automatic measure fetal head circumference in ultrasound images using convolutional neural networks | |
| Sun et al. | Virtual reality of recognition technologies of the improved contour coding image based on level set and neural network models | |
| CN111612770B (en) | Active screening-based focus detection system of semi-supervised focus detection network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |