Map information generation method, determination method, and program
阅读说明：本技术 地图信息生成方法、判定方法以及程序 (Map information generation method, determination method, and program ) 是由 有村秀孝 靳泽 平川和弥 于 2018-05-10 设计创作，主要内容包括：一种地图信息生成方法,其包括：细线化患者图像生成步骤,生成脑动脉瘤患者的脑血管的坐标信息与脑动脉瘤的坐标信息分别建立对应、且该脑血管被细线化的细线化患者图像信息；坐标对应建立步骤,将细线化患者图像信息所示的被细线化的脑血管的坐标、与脑血管被细线化的细线化参考图像所示的脑血管的参考坐标建立对应；以及地图信息生成步骤,基于针对多个脑动脉瘤患者的所述细线化患者图像信息的坐标对应建立步骤中建立起对应的脑血管的坐标及坐标信息,生成示出脑动脉瘤的坐标相对于参考坐标的发生概率的概率性地图信息。(A map information generation method, comprising: a thinning patient image generation step of generating thinning patient image information in which the coordinate information of a cerebral blood vessel of a cerebral aneurysm patient and the coordinate information of a cerebral aneurysm are respectively associated and the cerebral blood vessel is thinned; a coordinate correspondence establishing step of establishing correspondence between the coordinates of the thinned cerebrovascular shown in the thinned patient image information and the reference coordinates of the cerebrovascular shown in the thinned cerebrovascular reference image; and a map information generation step of generating probabilistic map information showing the probability of occurrence of the coordinates of the cerebral aneurysm with respect to the reference coordinates, based on the coordinates of the cerebral vessels and the coordinate information established in the coordinate correspondence establishing step of the thinned patient image information for the plurality of cerebral aneurysm patients.)
1. A map information generation method, comprising:
a thinning patient image generation step of generating thinning patient image information in which the coordinate information of a cerebral blood vessel of a cerebral aneurysm patient and the coordinate information of a cerebral aneurysm are respectively associated and the cerebral blood vessel is thinned;
a coordinate correspondence establishing step of associating coordinates of the thinned cerebrovascular shown in the thinned patient image information generated in the thinned patient image generating step with reference coordinates of the cerebrovascular shown in a thinned reference image in which the cerebrovascular is thinned; and
and a map information generation step of generating probabilistic map information showing the probability of occurrence of the coordinates of the cerebral aneurysm with respect to the reference coordinates, based on the coordinates of the cerebral vessels and the coordinate information established in the coordinate correspondence establishing step of the thinned patient image information for the plurality of cerebral aneurysm patients.
2. The map information generation method according to claim 1, wherein in the map information generation step, the probabilistic map information is generated based on a size of the cerebral aneurysm.
3. A method of determining, comprising:
a determination target thinned patient image generation step of generating determination target thinned patient image information in which coordinate information of a cerebral blood vessel of a cerebral aneurysm patient as a determination target is associated with coordinate information of a cerebral aneurysm, respectively, and the cerebral blood vessel is thinned;
a probabilistic map information generating step of generating the probabilistic map information by the map information generating method according to claim 1 or 2;
a2 nd coordinate association step of associating coordinates of the thinned cerebrovascular vessel shown in the probabilistic map information generated in the probabilistic map information generation step with reference coordinates of the cerebrovascular vessel shown in the judgment target thinned patient image information;
a2 nd map information generation step of generating 2 nd probabilistic map information based on the coordinates and coordinate information of the cerebral vessels corresponding to the coordinates and coordinate information established in the 2 nd coordinate correspondence establishment step; and
a false positive determination step of determining a false positive included in the 2 nd probabilistic map information by bayesian estimation.
4. The determination method according to claim 3, wherein, in the false positive determination step, a ratio of eigenvalues of a blackplug matrix representing a quadratic form of an ellipsoid when the cerebral aneurysm is regarded as the ellipsoid is used.
5. A program for causing a computer to execute the steps of:
a thinning patient image generation step of generating thinning patient image information in which the coordinate information of a cerebral blood vessel of a cerebral aneurysm patient and the coordinate information of a cerebral aneurysm are respectively associated and the cerebral blood vessel is thinned;
a coordinate correspondence establishing step of associating coordinates of the thinned cerebrovascular shown in the thinned patient image information generated in the thinned patient image generating step with reference coordinates of the cerebrovascular shown in a thinned cerebrovascular reference image; and
and a map information generation step of generating probabilistic map information showing the probability of occurrence of the coordinates of the cerebral aneurysm with respect to the reference coordinates, based on the coordinates of the cerebral vessels and the coordinate information established in the coordinate correspondence establishing step of the thinned patient image information for the plurality of cerebral aneurysm patients.
Technical Field
The present invention relates to a map information generation method, a map information determination method, and a map information generation program.
The application requests the priority of Japanese patent application No. 2017, Japanese patent application No. 2017-106643, which is filed on 30.5.7.2017, and the disclosure of the priority is cited.
Background
The prevalence of intracranial aneurysms (hereinafter referred to as cerebral aneurysms) is estimated to be between 3.6% and 6.0% based on anatomical predictions and angiographic diagnosis. Cerebral aneurysm refers to a disease of cerebrovascular disorder in the brain, which may cause local swelling or saccularization of blood vessels. Rupture of a cerebral aneurysm can lead to subarachnoid and intracranial hemorrhage. The mortality rate for subarachnoid or intracranial hemorrhage is 40% to 50%.
In recent years, high attention has been paid to prevention of stroke and dementia, and medical health diagnostic systems for intracranial diseases and cerebrovascular diseases have been established. In japan, medical health diagnosis called brain health examination is widely used. By 2009, a national survey showed that brain health checks could be diagnosed in 235 medical institutions certified in japan. Cerebral aneurysms were detected in 2.6% of the subjects by brain health examination.
A standard detection method used in brain aneurysm screening is Magnetic Resonance Angiography (MRA). MRA is used for vascular imaging to detect cerebral aneurysms, stenosis, occlusions or other cerebrovascular abnormalities. MRA is widely used because it has advantages of low toxicity, non-invasion, and the like, compared to an angiography method or a catheter angiography method using Computed Tomography (CT), because it does not use a contrast medium.
In the Maximum Intensity Projection (MIP) used in MRA image processing, there are sometimes cases where images overlap, or a cerebral aneurysm may be located in an unusual position. Therefore, it is difficult for a radiologist to detect a cerebral aneurysm by manual operation.
To assist radiologists in detecting asymptomatic unbroken cerebral aneurysms using MRA images, a framework of many Computer-aided diagnosis (CAD) systems has been developed. According to the study, it was statistically shown that: significant improvements can be seen in improving the diagnostic performance of less experienced radiologists and in reducing image reading time using CAD systems.
Further, an aneurysm measurement method, an aneurysm measurement device, and a computer program are known, in which an inflow side blood vessel and an outflow side blood vessel of a parent blood vessel of an aneurysm are designated and marked, and a blood vessel core line is extracted by a thinning process to specify an abnormality such as blood vessel dilatation or the like (for example, see patent document 1).
Disclosure of Invention
Problems to be solved by the invention
When determining the location of a cerebral aneurysm, a map showing the probability of occurrence of the cerebral aneurysm may sometimes be utilized. The map generation requires a professional to manually generate the map from the database, and this requires a long time. In the method of determining the location of a cerebral aneurysm using a map showing the probability of occurrence of the cerebral aneurysm, a technique of automatically generating a map showing the probability of occurrence of the cerebral aneurysm is required.
The present invention has been made in view of the above-described problems, and provides a map information generation method, a determination method, and a program that can automatically generate a map showing a probability of occurrence of a cerebral aneurysm.
Means for solving the problem
(1) The present invention has been made to solve the above-mentioned problems, and an aspect of the present invention provides a map information generating method including: a thinning-patient-image generating step of generating thinning-patient-image information in which the coordinate information of a cerebral blood vessel of a cerebral aneurysm patient and the coordinate information of a cerebral aneurysm are respectively associated with each other and the cerebral blood vessel is thinned; a coordinate correspondence establishing step of associating coordinates of the thinned cerebrovascular shown in the thinned patient image information generated in the thinned patient image generating step with reference coordinates of the cerebrovascular shown in a thinned cerebrovascular reference image; and a map information generation step of generating probabilistic map information showing the probability of occurrence of the coordinates of the cerebral aneurysm with respect to the reference coordinates, based on the coordinates of the cerebral vessels and the coordinate information established in the coordinate correspondence establishing step of the thinned patient image information for the plurality of cerebral aneurysm patients.
(2) In addition, according to an aspect of the present invention, in the map information generation method, the probabilistic map information is generated based on a size of the cerebral aneurysm in the map information generation step.
(3) Further, an aspect of the present invention provides a determination method including: a determination target thinned patient image generation step of generating determination target thinned patient image information in which coordinate information of a cerebral blood vessel of a cerebral aneurysm patient as a determination target is associated with coordinate information of a cerebral aneurysm, respectively, and the cerebral blood vessel is thinned; a probabilistic map information generating step of generating the probabilistic map information by the map information generating method according to
(4) In addition, in one aspect of the present invention, in the above determination method, the false positive determination step uses a eigenvalue ratio of a quadratic blackplug matrix that represents an ellipsoid when the cerebral aneurysm is regarded as the ellipsoid.
(5) Further, an aspect of the present invention provides a program for causing a computer to execute the steps of: a thinning-patient-image generating step of generating thinning-patient-image information in which the coordinate information of a cerebral blood vessel of a cerebral aneurysm patient and the coordinate information of a cerebral aneurysm are respectively associated with each other and the cerebral blood vessel is thinned; a coordinate correspondence establishing step of associating coordinates of the thinned cerebrovascular shown in the thinned patient image information generated in the thinned patient image generating step with reference coordinates of the cerebrovascular shown in a thinned cerebrovascular reference image; and a map information generation step of generating probabilistic map information showing the probability of occurrence of the coordinates of the cerebral aneurysm with respect to the reference coordinates, based on the coordinates of the cerebral vessels and the coordinate information established in the coordinate correspondence establishing step of the thinned patient image information for the plurality of cerebral aneurysm patients.
Effects of the invention
According to the present invention, a map showing the probability of occurrence of a cerebral aneurysm can be automatically generated.
Drawings
Fig. 1 is a schematic diagram of an example of a distribution of cerebral aneurysms and false positives in a cerebrovascular image of an embodiment.
Fig. 2 is a schematic diagram of one example of the automatic creation process of the cerebral aneurysm probability map (atlas) of the present embodiment.
Fig. 3 is a schematic diagram showing an example of a thinned image of a main blood vessel as a reference and a thinned image of a main blood vessel of a clinical case before registration according to the present embodiment.
Fig. 4 is a schematic diagram showing an example of a thinned image of a main blood vessel as a reference and a thinned image of a main blood vessel of a clinical case after registration according to the present embodiment.
Fig. 5 is a schematic diagram of an example of a cerebral aneurysm probability map of the present embodiment.
Fig. 6 is a schematic diagram of an example of a bayesian CAD framework using a cerebral aneurysm probability map according to the present embodiment.
Fig. 7 is a diagram showing an example of the distribution of the ratio of false positives to the feature value of a cerebral aneurysm according to the present embodiment.
Fig. 8 is a schematic diagram of an example of a cerebral aneurysm probability map of the present embodiment viewed from the front to the rear.
Fig. 9 is a schematic diagram of an example of the FROC curve of the bayesian CAD framework of the present embodiment.
Fig. 10 is a schematic diagram showing an example of a functional structure of a bayesian CAD framework using a cerebral aneurysm probability map according to the present embodiment.
Detailed Description
(embodiment mode)
Embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
Fig. 1 is a schematic diagram of an example of a distribution of cerebral aneurysms and false positives in a cerebrovascular image of an embodiment. The cerebrovascular image is an image of a cerebral vessel when the brain is viewed from the front.
In routine diagnosis of CAD systems, various methods have been developed for detecting cerebral aneurysms. As an example, an elliptical Convex region enhancement (ECE) filter capable of reducing false positives in curved portions of blood vessels and selectively enhancing the shape of cerebral aneurysms is proposed. By using the ECE filter, the number of false positives was reduced to 51.4%.
However, as shown in fig. 1, some false positives are detected at the entrance portion FP1 of the image area of the imaged blood vessel V1, or at the extremities FP2, FP3 of the imaged blood vessel V1. In addition, a true positive was detected at position a 0.
To further reduce false positives, it is useful to create a map, i.e. a probability map, showing the probability of occurrence of a cerebral aneurysm, based on past diagnosis and prior information on the location of the cerebral aneurysm from existing databases. Most cerebral aneurysms are known to occur around the cerebral arterial loop.
Probability maps are used in the fragmentation of organs such as the liver and in brain structures. However, methods using a priori information of the common location where a cerebral aneurysm occurs are not well known.
In this embodiment, a method for automatically creating a probability map of a cerebral aneurysm from a database using a computer is provided. In a procedure using a computer, a priori information about the location of the cerebral aneurysm will be effectively utilized.
The location of a cerebral aneurysm depends on the structure of blood vessels, which have a common pattern, but the shape of blood vessels varies from individual to individual. It is difficult to register the reference data with multiple clinical data using only rigid registration. In the method, for the registration between main vessels of cerebral vessels, a Thin Plate Spline function-Robust Point Matching (TPS-RPM) which is a non-rigid registration method is used.
In the present embodiment, the position information of the cerebral aneurysm obtained from 89 unbroken cerebral aneurysms of 72 patients is used to create a probability map of the cerebral aneurysm. In the present embodiment, a CAD framework for reducing the number of false positives is operated based on a bayesian estimation combining a probability map of a cerebral aneurysm and an ECE filter.
The probability map of a cerebral aneurysm in the present embodiment is referred to as a probability map PA.
(clinical cases)
The database of this embodiment includes data for 72 patients, and the total number of patients had 89 unbroken cerebral aneurysms. Of these, 72 patients consisted of 17 males and 55 females between the ages of 36 and 89. The database includes data of 1 healthy person as reference data. The reference data is data for women aged 45 who received MRA diagnosis between 2006 and 2007, or between 2010 and 2014. The reference data was selected from data of 30 persons who did not suffer from a cerebral aneurysm with reference to the shape of a blood vessel of a healthy person.
The cerebral aneurysms contained in the database were confirmed by two experienced neuroradiologists using a CT angiography method or a Digital subtraction angiography method (DSA) according to the japan society of cerebral apoplexy guidelines. MRA images of the patient were acquired using a 3.0Tesla magnetic resonance image scanner. In a three-dimensional MRA image, each case contains 112 to 172 slices. Wherein the slice is a slice composed of a thickness of 1 mm to 1.2 mm and a distance of 0.5 mm to 0.7 mm. The slice image is 512 x 512 pixels with a pixel size of 0.3516 mm. The original three-dimensional MRA image is converted using three-dimensional interpolation into isotropic volume data consisting of matrices of dimensions 512 x (224-258) and isotropic voxels of dimensions 0.3516 mm.
The major diameter and the two minor diameters of the cerebral aneurysm are measured by manual operation along the major axis, the medial axis, and the minor axis of each ellipsoid, respectively. Wherein the major axis, the central axis and the minor axis of the ellipsoid are orthogonal to each other. The measurement was performed twice by each of two measuring staff on the MRA image using Multi-Planar Reconstruction (MPR) software. The major diameter of 89 cerebral aneurysms ranged from 1.4 mm to 10.6 mm, with an average of 4.4 mm.
The positional distribution of cerebral aneurysms in the present embodiment will be described.
40% of Cerebral aneurysms occur in the Middle Cerebral Artery (MCA: Middle Cerebral Artery), and 38% in the Internal Carotid Artery (ICA: Internal cardiac Artery) and the Posterior Carotid Communicating Artery (ICA-Pcom: ICA-Posterior Communicating Artery). Thus, most cerebral aneurysms occur within MCA, ICA and ICA-Pcom.
10% of cerebral aneurysms occur in the Anterior Communicating arteries (Acom: antioxidant Communicating Artery). 5% of Cerebral aneurysms occur in the Anterior Cerebral Artery (ACA: antioxidant coronary Artery). 3% of cerebral aneurysms occur in the distal Basilar Artery (BAtip: Tip of Basilar Artery) or the branching of the Basilar-Superior cerebellar Artery (BA-SCA: Basilar Artery-Superior cerebellar Artery). 2% of cerebral aneurysms occur in the posterior and inferior vertebral artery-cerebellum (VA-PICA: vertebral arteries-porterior afferior cerebellar arteries) bifurcation.
(creation of cerebral aneurysm probability map)
The probability map PA is determined in the blood vessel image as a reference by using a gaussian sphere showing the location and size of the cerebral aneurysm of the database. The position of each cerebral aneurysm in the blood vessel image as a reference is determined by converting the position of the cerebral aneurysm of the clinical image into the position of the blood vessel image as a reference using the deformation vector field. Here, the deformation vector field is generated by performing registration of a thinned image of the main vessels between the clinical case and the vessel image as a reference using the TPS-RPM method. The gaussians were then deployed at the location of the cerebral aneurysm, which was over-registered. Here, the standard deviation of the gaussian distribution of the gaussian sphere is one quarter of the value of the major axis of each cerebral aneurysm.
Fig. 2 is a flowchart showing a process flow of creating a probability map PA from a database. In the present embodiment, the method of creating the probability map PA is referred to as a map information generation method M.
(step S10) the thinning image of the main cerebral vessels is generated from the MRA image.
To perform robust registration of the vessels, a thinned image of the major vessels is extracted. Here, the reason for using the main blood vessels is that the main blood vessels are surrounded by the common location where the cerebral aneurysm occurs, and the distribution of the main blood vessels is less indefinite than the distribution of the branch blood vessels.
Based on the information of the signal intensity of the MRA image, the main blood vessels are extracted. First, the maximum voxel value is selected from a spherical region with a radius of 30 mm in the center of the MRA image. Then, the initial vessel region included in the spherical region is distinguished with the signal intensity of 60% of the maximum voxel value as a threshold. The volume ratio increase was observed and a regional dilation method was applied to the initial vessel region. Thus, a threshold value relative to the signal strength is automatically determined, and a region relative to the main blood vessels is acquired from the original MRA image.
And performing thinning processing on the acquired images of the main blood vessels. Thinning of the blood vessel image is performed, for example, using ImageJ software. Morphological processing was performed as follows.
To remove the lumen within the vessel, a closing (closing) process is performed. The closing process is a process in which the contraction process is repeated after the expansion process. The dilation process is a process of replacing the maximum signal intensity in the vicinity of a voxel of interest with the signal intensity of the voxel. The contraction processing is processing for replacing the minimum signal intensity in the vicinity of a voxel of interest with the signal intensity of the voxel.
In order to remove the branch vessels, a constriction treatment is performed. And (3) applying a three-dimensional thinning algorithm to generate a thinning image of the main blood vessel.
In step S10, a thinned image SV2 of the clinical case is generated. The generated thinned image SV2 is generated by associating the coordinate information of the cerebral blood vessel of the cerebral aneurysm patient with the coordinate information of the cerebral aneurysm, and generating thinned
A thinned image SV1 is generated as a reference. The generated thinned image SV1 is generated by associating the coordinate information of the cerebral blood vessel of the cerebral aneurysm patient with the coordinate information of the cerebral aneurysm, and generating thinned
(step S20) the position of the cerebral aneurysm of the clinical case is registered to the position as the reference by registering the thinned image of the clinical case with the thinned image as the reference.
In order to create an atlas of a cerebral aneurysm, non-rigid registration of the transformation from the location of the cerebral aneurysm of a clinical case to the location of a thinned image as a reference needs to be performed.
First, non-rigid registration is performed between a point set of a thinned image of a main blood vessel as a reference and a point set of a thinned image of a main blood vessel of a clinical case using the TPS-RPM method. Thereby, a deformation vector field defined by each point is obtained.
Next, the location of the cerebral aneurysm of the clinical case is registered to a location on the thinned image of the main vessel as a reference, using the deformation vector at the closest point of the thinned image of the main vessel as a reference.
In this manner, in step S20, the coordinates of the thinned cerebral vessels shown in the thinned image information SVI2 of the clinical case generated in step S10 are associated with the reference coordinates of the cerebral vessels shown in the thinned reference image in which the cerebral vessels are thinned, shown in the thinned image information SVI1 as a reference.
FIG. 3 is a schematic diagram of one example of a thinned image SV1-1 of a main blood vessel as a reference before registration with a thinned image SV2 of a main blood vessel of a clinical case. In the thinned image SV1-1 and the thinned image SV2, the position of the blood vessel is deviated.
Fig. 4 is a schematic diagram of one example of a thinned image SV1-2 of a main blood vessel as a reference and a thinned image SV3 of a main blood vessel of a clinical case after registration is performed. In the thinned image SV1-2 and the thinned image SV2, the positions of the blood vessels correspond to each other, and the deviation is smaller than that in fig. 3.
Returning to fig. 2, the description of the process of creating the probability map PA is continued.
(step S30) a probability map PA is created from the registered cerebral aneurysm.
By the registration in step S20, the position of the cerebral aneurysm on the thinned image of the main blood vessel as a reference is acquired. It is assumed that the probability of occurrence of a cerebral aneurysm around the cerebral aneurysm location follows a normal distribution. The probability map PA is created based on a gaussian distribution expressed by equation (1) around the position of the cerebral aneurysm where registration is completed.
[ equation 1]
Wherein the position vector v is a three-dimensional position vector relative to the voxel position, the probability Pa (v) is the probability that the voxel at the position shown by the position vector v is contained in the cerebral aneurysm, and the position vector c isiAs a position vector relative to the position of the ith cerebral aneurysm, the value σiThe standard deviation of the gaussian distribution (one quarter of the length of the major diameter of the ith cerebral aneurysm).
In this manner, in step S30, based on the coordinates and the coordinate information of the corresponding cerebral vessels established in the coordinate correspondence establishing step (step S20) of the thinned image information SVI2 for a plurality of clinical cases, a probability map PA showing the probability of occurrence of the coordinates of the cerebral aneurysm with respect to the coordinates of the cerebral vessels shown in the thinned image information SVI1 as a reference is generated.
Then, in step S30, a probability map PA is generated based on the major diameter length of the cerebral aneurysm.
Fig. 5 is a schematic diagram of an example of the probability map PA of the present embodiment. Probability maps PA were created based on 89 cerebral aneurysms. The probability map PA is drawn in the direction of observing the main vessels of the brain from the front.
On the thinned image SV1 of the main blood vessel as a reference, a probability distribution of a cerebral aneurysm is shown. For example, brain aneurysms occur more at positions a1, a2, A3, a4, a5, a6, and a7, corresponding to more location distribution of brain aneurysms in clinical cases. On the other hand, for example, a cerebral aneurysm hardly occurs at the position R1 corresponding to the little distribution of the positions of cerebral aneurysms of clinical cases.
(CAD frame)
A CAD framework using bayesian estimation based on probability atlas PA is illustrated.
Fig. 6 is a schematic diagram of a CAD framework using bayesian estimation based on probability atlas PA. Hereinafter, the CAD framework using the bayesian estimation is referred to as a bayesian CAD framework.
In step S110, an MRA image DI of one case for determination is selected and acquired from the MRA images for clinical cases.
In step S120, an ECE filter is applied to the MRA image DI acquired in step S110. In step S120, the image to which the ECE filter is applied and the eigenvalue ratio of the black plug matrix representing the quadratic form of an ellipsoid when each candidate cerebral aneurysm is regarded as an ellipsoid are acquired. Selectively enhancing the shape of the candidate cerebral aneurysm in the image after applying the ECE filter.
In step S130, the coordinate information of the cerebral blood vessel of the cerebral aneurysm patient to be determined is associated with the coordinate information of the cerebral aneurysm, and the thinned image information SVI4 to be determined in which the cerebral blood vessel is thinned is generated. Here, the determination target thinned image indicated by the determination target thinned image information SVI4 is referred to as a thinned image SV 4.
In step S140, a probability map PA is generated by the map information generation method M shown in steps S10, S20, and S30.
In step S150, a probability map PA-1 is generated by registering the probability map PA generated in step S140 to the thinned image SV 4. This registration is performed in the same manner as the registration of the thinned image SV2 of the main blood vessel of the clinical case to the thinned image SV1 of the main blood vessel as a reference, which is shown in step S20. The locations of the individual cerebral aneurysms, indicated by gaussians, were registered to the thinned image SV4 by using the deformation vector field. Here, the deformation vector is acquired by registration of the blood vessel image by the TPS-RPM method.
Cerebral aneurysms usually occur on the surface of blood vessels. Therefore, local position adjustment of the gaussian sphere of the probability map PA is required. In the thinned image SV4, the probability map PA-1 is acquired by moving each gaussian sphere to the nearest blood vessel surface.
In step S160, bayesian estimation based on the probability map PA-1 and the feature value ratios obtained from the ECE filter is applied to the candidate cerebral aneurysm. This can reduce the number of false positives in the MRA image DI of the case for determination.
The ECE filter is designed based on the eigenvalue of a quadratic black matrix representing an ellipsoid in a voxel indicated by a position vector v. The ratio of the maximum to minimum of the eigenvalues corresponds to the ratio of the minor axis to the central axis of the ellipsoid when modeling the shape of the local signal strength by the ellipsoid. The corresponding relationship is shown in
[ equation 2]
Wherein, the
The eigenvalue ratio is used as likelihood in bayesian estimation.
Fig. 7 is a graph showing the distribution of eigenvalue ratios relative to false positives and cerebral aneurysms. The distribution RD1 of the eigenvalue ratio with respect to the cerebral aneurysm is a normal distribution expressed by equation (3).
[ equation 3]
Wherein the mean value μTPIs a characteristic value ratio I (v)TP) Mean value of (3), variance σTP 2Is a characteristic value ratio I (v)TP) The variance of (c). Probability PTP(I(νTP)|νTP) Is when the position vector vTPThe eigenvalue ratio is the ratio I (v) when the location shown is contained within a cerebral aneurysmTP) Probability (likelihood).
The distribution RD2 of the eigenvalue ratio relative to the false positives is a normal distribution represented by equation (4).
[ equation 4]
Wherein the mean value μFPAs a characteristic value ratio I (v)FP) Mean value of (3), variance σFP 2As a characteristic value ratio I (v)FP) The variance of (c). Probability PFP(I(νFP)|νFP) Is when the position vector vTPThe eigenvalue ratio is the ratio I (v) when the indicated location is not contained within a cerebral aneurysmFP) Probability (likelihood).
Returning to fig. 6, the description of the bayesian CAD framework is continued.
When the feature value ratio I (v) is obtained, the posterior probability that the voxel represented by the position vector v is located inside the cerebral aneurysm is given by the probability P (v | I (v)). The probability P (ν | I (ν)) is given by equation (5).
[ equation 5]
Wherein the probability Pa(v) is the probability of occurrence of a cerebral aneurysm given by equation (1).
Based on the occurrence probability P (ν | I (ν)) of the cerebral aneurysm given by equation (5) acquired in step S160, the probability distribution of the cerebral aneurysm of the probability map PA-1 is updated, and the probability map PA-2 is generated. False positives were determined using a probability map PA-2.
In step S170, a false positive is determined based on a rule reasoning and Support Vector Machine (SVM).
For evaluation, Leave-One-Out (Leave-One-Out) cross-validation was used. One of 89 clinical cases was selected as a case for determination, and the remaining 88 clinical cases were used as clinical cases, and a bayesian CAD framework was performed. This operation was repeated a total of 89 times until all 89 clinical cases were selected as
(results)
Fig. 8 is a schematic diagram showing an example of an image obtained by the maximum intensity projection method when the probability map PA-2 of the present embodiment is viewed from the front to the rear.
In the results obtained in this example, most of cerebral aneurysms were produced in the blood vessel V3 around the position a12 (cerebral artery loop), the positions a8, a11(MCA), and the positions a9, a10(ICA curved portion). On the other hand, for example, a cerebral aneurysm hardly develops at the position R2.
The results reproduce data showing the proportion of cerebral aneurysms in size and location in a non-ruptured cerebral aneurysm prevalence study.
Fig. 9 is a schematic diagram of an example of the FROC curve of the bayesian CAD framework of the present embodiment. Here, the FROC curve is a curve showing a relationship between false positives and sensitivity.
A free-response receiver operating characteristic (FROC) curve G1 when using bayesian estimation based on probability maps and eigenvalue ratios obtained from ECE filters is compared to a FROC curve G2 when not in use.
According to the present embodiment, when 89 cases were used, the number of false positives was compared among the results with the highest sensitivity. It has been demonstrated that by incorporating a probability map into the bayesian CAD framework, the initial value of false positives is reduced from 25.2 to 7.2 with a sensitivity of 93.1%. And it was confirmed that: the average number of false positives per case was reduced from 19 to 5.4 with a sensitivity of 80%.
Fig. 10 is a schematic diagram showing an example of a functional structure of a bayesian CAD framework using the probability map according to the present embodiment. The bayesian CAD frame of this embodiment is referred to as frame F.
The framework F includes the cerebral aneurysm
The cerebral aneurysm
The determination
The feature value
The thinned
And the probability map
The clinical
The reference
The PA thinning-out
The PA thinned-
The PA line-thinned
The PA coordinate
The PA coordinate
The map
The map
The map
The coordinate
The coordinate
The 1 st false
The 1 st false
The 2 nd false
The
(conjugal language)
In rigid registration and brain cell-based registration, although there is a limitation on different positions of blood vessels anatomically, according to the map information generation method M, non-rigid registration by a blood vessel-based thinned image can be overcome.
When creating the probabilistic atlas, registration of the location of the cerebral aneurysm to the main vessel as a reference can be done without manual registration. Therefore, according to the creation method, the opportunity can be brought to potential users. For example, in a research institute or hospital, by utilizing a priori information about the location of a cerebral aneurysm, the owned database may be more practically and efficiently utilized.
Compared with the manual method in the prior art, the computer alignment method provided by the embodiment reduces the operation amount of the user and reduces the difference generated by the observer. The time required for the registration of blood vessels between the thinned image of the clinical case and the thinned image as a reference was 1 minute 46.2 ± 3.7 seconds in the case of using an ordinary personal computer (clock number 3.2GHz, core number 6 nuclei). This is a number that can tolerate daily use.
The CAD framework provided in this embodiment can detect 93.1% of all cerebral aneurysms with 7.2 false positives per case, or can detect 80% of all cerebral aneurysms with 5.4 false positives per case.
Compared to the 1.5 tesla magnetic resonance scanner, the 3 tesla magnetic resonance scanner requires only half the scan time, making the scan more comfortable and reducing the offset caused by movement.
The map information generation method M according to the above embodiment includes: a thinning-out patient image generation step (step S10) of generating thinning-out patient image information (thinning-out image information SVI2) in which the coordinate information of the cerebral blood vessels of a cerebral aneurysm patient and the coordinate information of the cerebral aneurysm are correlated with each other and the cerebral blood vessels are thinned out; a coordinate association step (step 20) of associating the coordinates of the thinned cerebral vessels shown by the thinned patient image information (thinned image information SVI2) generated in the thinned patient image generation step (step S10) with the reference coordinates of the cerebral vessels shown by the thinned cerebral vessel reference images (thinned images SV1-1 and SV 1-2); and a map information generation step (step S30) of generating probabilistic map information (probability spectrum PA) showing the probability of occurrence of the coordinates of the cerebral aneurysm with respect to the reference coordinates (the coordinates of the cerebral vessels shown by the thinned-out image information SVI1 as a reference) based on the coordinates and the coordinate information of the cerebral vessels corresponding to the coordinate correspondence establishing step (step S20) of the thinned-out patient image information (thinned-out image information SVI2) for the plurality of cerebral aneurysm patients. Therefore, according to the map information generation method M, a map showing the probability of occurrence of a cerebral aneurysm can be automatically generated.
According to the map information generation method M provided in the above embodiment, probabilistic map information (probability map PA) is generated based on the size of a cerebral aneurysm (the length of the major axis of the cerebral aneurysm). Therefore, according to the map information generation method M, information on the size of a cerebral aneurysm can be reflected in a map showing the probability of occurrence of the cerebral aneurysm.
The judgment method provided by the above embodiment includes: a determination target thinning-out patient image generation step (step S130) of generating determination target thinning-out patient image information (determination target thinning-out image information SVI4) in which the coordinate information of the cerebral blood vessel of the cerebral aneurysm patient as the determination target and the coordinate information of the cerebral aneurysm are correlated with each other and the cerebral blood vessel is thinned out; a probabilistic map information generation step (step S140) of generating probabilistic map information (probability map PA) by a map information generation method M; a2 nd coordinate association step (step S150) of associating the coordinates of the thinned cerebral vessels shown in the probabilistic map information (probability map PA) generated in the probabilistic map information generation step (step S140) with the reference coordinates of the cerebral vessels (the coordinates of the cerebral vessels shown in the judgment target thinned image information SVI4) shown in the judgment target thinned patient image information (judgment target thinned image information SVI 4); a2 nd map information generation step (step S150) of generating 2 nd probabilistic map information (probability map PA-1) based on the coordinates and coordinate information of the cerebral vessels corresponding to each other established in the 2 nd coordinate correspondence establishment step (step S150); and a false positive determination step of determining false positives included in the 2 nd probabilistic map information (probability map PA-1) by bayesian estimation. Therefore, according to the determination method provided in the above embodiment, the number of false positives in the determination result can be reduced.
According to the determination method provided in the above embodiment, the eigenvalue ratio of the blackplug matrix of the quadratic form representing an ellipsoid is used when the cerebral aneurysm is regarded as the ellipsoid. According to the determination method provided in the above-described embodiment, the number of false positives in the determination result can be reduced based on the shape of the cerebral aneurysm.
While the embodiments of the present invention have been described in detail with reference to the drawings in the specification, the specific configurations are not limited to these embodiments, and may be changed as appropriate without departing from the spirit of the present invention. The structures described in the above embodiments may be combined.
In addition, the invention can also be applied to the creation of a probability map of blood vessels, air pipes and alveoli aiming at the parts with the tendency to cause lung cancer. In this case, for example, a probability map can be created by associating the major blood vessels of the cerebral blood vessels described in the above embodiment with blood vessels or trachea. As described above, the present invention can be applied by registering the position of the affected part with the organ as a reference, as long as the organ has a common structure among individuals. The invention may also be applied to organs such as bones, the major nervous system, the major lymphatic vessels, the digestive tract, etc.
In addition, each part included in each apparatus in the above-described embodiments may be realized by dedicated hardware, or may also be realized by a memory and a microprocessor.
Each part included in each device is configured by a memory and a CPU (central processing unit), and the functions thereof can be realized by loading and executing a program for realizing the functions of each part included in each device into the memory.
Further, a program for realizing the functions of the respective sections included in the respective apparatuses may be recorded in a computer-readable storage medium, and the processing of the respective sections included in the control section may be executed by causing a computer system to read and execute the program recorded in the storage medium. The "computer system" referred to herein includes hardware such as an OS and peripheral devices.
Further, the "computer system" also includes a homepage providing environment (or display environment) if the WWW system is utilized.
The term "computer-readable storage medium" refers to portable media such as flexible disks, magneto-optical disks, ROMs, and CD-ROMs, and storage devices such as hard disks built in computer systems. Further, the "computer-readable storage medium" means a medium including a medium that dynamically holds a program for a short time, such as a communication line that transmits the program via a network such as the internet or a communication line such as a telephone line, and a medium that holds the program for a certain time in this case, such as a volatile memory inside a computer system serving as a server or a client. The program may be a program for realizing a part of the above functions, or may be a program for realizing the above functions by combining with a program already recorded in a computer system.
[ description of reference ]
A1 … brain aneurysm position estimating device, a2 … image supplying unit, A3 … presenting unit, a10 … determination image acquiring unit, a11 … feature value ratio calculating unit, a12 … thinned image generating unit, a 13 … probability map automatic generating unit, a 14 … coordinate correspondence establishing unit, a 15 … 1 st false positive determining unit, a 16 … 2 nd false positive determining unit, a 130 … clinical image acquiring unit, a 131 … reference image acquiring unit, a 132 … PA thinned image generating unit, a 133 … PA coordinate correspondence establishing unit, a 134 … map information generating unit, a V1 … blood vessel, an FP1 … inlet portion, FP2, an FP3 … terminal, a1, a2, A3, a4, A5, A6, a7, R1 … position, SV1-1, SV1-2, SV2, and SV3 … thinned images.
- 上一篇：一种医用注射器针头装配设备
- 下一篇：用于控制行动装置的方法和装置