CN115661132A - Computer storage medium and device for detecting benign and malignant pulmonary nodules - Google Patents
Computer storage medium and device for detecting benign and malignant pulmonary nodules Download PDFInfo
- Publication number
- CN115661132A CN115661132A CN202211552686.6A CN202211552686A CN115661132A CN 115661132 A CN115661132 A CN 115661132A CN 202211552686 A CN202211552686 A CN 202211552686A CN 115661132 A CN115661132 A CN 115661132A
- Authority
- CN
- China
- Prior art keywords
- lung
- dimensional
- benign
- nodule
- malignant
- 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.)
- Pending
Links
Images
Landscapes
- Apparatus For Radiation Diagnosis (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to the field of medical imaging techniques, and provides a computer storage medium having program instructions embodied therein for detecting benign and malignant lung nodules for causing a corresponding processor to perform the steps of: acquiring CT images of the known benign and malignant condition of the pulmonary nodule and the benign and malignant condition of the pulmonary nodule to be detected; after the lung nodule characteristic vectors in the lung CT image are set, corresponding lung nodule characteristic vectors in the CT image of the known benign and malignant condition of the lung nodule and the benign and malignant condition of the lung nodule to be detected are respectively extracted; performing clustering analysis on the lung nodule characteristic vectors to obtain various clustering results, and generating a reference clustering data set through the various clustering results; and detecting the benign and malignant condition of the pulmonary nodule to be detected by using the reference cluster data set. The invention provides accurate quantitative analysis for doctors through the quantitative processing technology of images, thereby improving the repeatability of the diagnosis result of the doctors and the consistency of the image and the disease explanation to a certain extent, and improving the accuracy and the efficiency of the diagnosis of the doctor.
Description
Technical Field
The present application relates to the field of medical imaging technologies, and in particular, to a computer storage medium and an apparatus for detecting benign and malignant pulmonary nodules.
Background
The discrimination and diagnosis of the benign and malignant lung nodules is the key for realizing the early diagnosis of the lung cancer. CT is considered as the best medical imaging means for examining lung diseases, but CT belongs to three-dimensional tomographic images, and has a large amount of image data, especially HRCT (High Resolution three-dimensional tomographic image), the amount of image data is increased sharply compared with the amount of data in a common CT image set, generally, one whole lung HRCT includes 200 to 500 scanned images, or even more, so that in a large-scale lung disease screening process, radiologists face a large number of CT images, and missed diagnosis and misdiagnosis of diseases are easily caused by diagnosis fatigue. Therefore, a computer aided diagnosis system for detecting benign and malignant lung nodules is needed to effectively assist a doctor in diagnosing benign and malignant lung nodules with suspicious lesions through a large number of lung CT images.
Disclosure of Invention
In view of the deficiencies in the prior art and the needs of practical applications, the present invention provides, in a first aspect, a computer storage medium for detecting benign and malignant lung nodules, the computer storage medium storing a computer program comprising program instructions that, when executed by a processor, cause the processor to perform the steps of: acquiring lung electron computer tomography images of known benign and malignant pulmonary nodules and benign and malignant pulmonary nodules to be detected; setting a lung nodule characteristic vector in a lung electron computed tomography image; extracting a first pulmonary nodule feature vector in a pulmonary electron computed tomography image of a known benign and malignant condition of a pulmonary nodule; obtaining a plurality of clustering results through the first lung nodule feature vector, and generating a reference clustering data set through the plurality of clustering results; extracting a second pulmonary nodule feature vector in the pulmonary electron computed tomography image of the benign and malignant condition of the pulmonary nodule to be detected; and detecting benign lung nodules and malignant lung nodules in the lung electron computed tomography image of the benign and malignant condition of the lung nodules to be detected by utilizing the second lung nodule feature vector and combining the reference cluster data set. The method extracts a plurality of characteristic vectors of a lung electron computed tomography image of the known benign and malignant condition of the lung nodule, generates a reference cluster data set for judging the benign and malignant condition of the lung nodule after cluster analysis, and identifies the extracted second lung nodule characteristic vector and diagnoses the benign and malignant condition of the lung nodule to be detected by using the reference cluster data set. The invention provides accurate quantitative analysis for doctors through the image quantization technology, overcomes the defects of inertia of human eyes and insensitivity to gray level, and can reduce the adverse effect of subjective factors of doctors on film reading results, thereby improving the repeatability of the diagnosis results of the doctors and the consistency of image and disease interpretation to a certain extent, and simultaneously reducing the omission of visual observation in the independent film reading diagnosis process of the doctors and the workload of film reading, thereby improving the accuracy and efficiency of the doctor diagnosis of the state of illness.
Optionally, the setting of the feature vector of the lung nodule in the lung electron computed tomography image includes the following steps: setting two-dimensional feature vectors in a single-layer lung electron computed tomography image, wherein the two-dimensional feature vectors comprise two-dimensional morphological feature vectors, two-dimensional gray-scale feature vectors and two-dimensional gradient feature vectors; and superposing a plurality of single-layer lung electron computed tomography images to form a lung nodule three-dimensional image, and setting a three-dimensional characteristic vector of the lung nodule three-dimensional image, wherein the three-dimensional characteristic vector comprises a three-dimensional morphological characteristic vector, a three-dimensional gray characteristic vector and a three-dimensional ladder characteristic vector.
Optionally, the three-dimensional grayscale feature vector includes a contrast-wise grayscale feature vector, the contrast-wise grayscale feature vector including:
wherein, inside represents the inner region of the three-dimensional lung nodule,
，outsideI represents the expanded peripheral region of the three-dimensional lung nodule,
and
represent the same class of features within the insert and outcidel regions respectively,
and
indicate the standard deviation corresponding to the same class of features in the inner and outer idei regions, respectively.
Optionally, the three-dimensional gradient feature vector comprises components of the gradient feature vector on three-dimensional coordinate axes, the components
The following formula is satisfied:
wherein, the first and the second end of the pipe are connected with each other,
，
representing the components of the three-dimensional gradient feature vector on the X-axis,
representing the component of the three-dimensional gradient feature vector on the Y-axis,
representing the component of the three-dimensional gradient feature vector in the Z-axis,
the gray value of the pixel point with the number i in the square field pixel of the central pixel point of the three-dimensional ladder characteristic vector is represented,
representing the weight of the pixel point numbered i,
in which
，
，
Representing the coordinates of a central pixel point
To the pixel point coordinate with number i
The euclidean distance between them.
Optionally, the obtaining multiple kinds of clustering results through the first lung nodule feature vector, and generating a reference clustering data set through the multiple kinds of clustering results includes the following steps: carrying out clustering analysis on the first lung nodule characteristic vector by utilizing different clustering algorithms to obtain various clustering results; or carrying out clustering analysis on the first lung nodule characteristic vector through the same clustering algorithm with different parameters and different initial values to obtain a plurality of clustering results; and combining the multiple clustering results to obtain a combined result, and carrying out clustering analysis on the combined result to obtain a reference clustering data set.
Further optionally, the different clustering algorithms include a Kmeans clustering method, a HC clustering algorithm, a MeanShift algorithm, and a Kmeans-HC combined clustering algorithm.
Further optionally, the combined result satisfies the following formula:
wherein the content of the first and second substances,
a matrix obtained by combining a plurality of kinds of clustering results is represented,
，
representation matrix
Middle sample
And samples
The corresponding numerical value at the indicated position is,
，
representing the total number of clustering results.
Optionally, the detecting benign pulmonary nodules and malignant pulmonary nodules in the lung electrical computed tomography image of the benign and malignant condition of the pulmonary nodules to be detected by using the second lung nodule feature vector in combination with the reference cluster data set includes the following steps:
obtaining pulmonary nodule feature vector x aiming at ith malignant feature subclass in reference cluster data set respectively
Bayes posterior probability of
And the jth benign feature subclass
Bayes posterior probability of
；
Setting a good/bad judgment boundary
The benign-malignant determination boundary satisfies the following equation:
converting the good and malignant judgment boundary formula into a Markov space linear judgment equation;
obtaining a Bayesian posterior probability of the second pulmonary nodule feature vector
And Bayes posterior probability
；
And preliminarily judging the benign and malignant degree of the second lung nodule feature vector according to a benign and malignant degree judgment boundary formula, wherein the benign and malignant degree judgment satisfies the following formula:
respectively calculating the second lung nodule feature vector to the jth benign feature subclass
Mahalanobis distance of center
And to the ithSubclass of malignancy characteristics
Mahalanobis distance of center
；
According to the Mahalanobis distance
And said Mahalanobis distance
Combined decision coordinates
At the mahalanobis location, the second lung nodule feature vector is again determined to be benign or malignant: if the second lung nodule feature vector belongs to the benign feature after the primary judgment, the judgment coordinate corresponding to the second lung nodule feature vector
If the second lung nodule feature vector is located below the mahalanobis space linear judgment straight line, the second lung nodule feature vector is false positive; if the second lung nodule feature vector is judged to belong to the malignant feature for the first time, judging coordinates corresponding to the second lung nodule feature vector
If the second pulmonary nodule feature vector is located above the mahalanobis space linear judgment straight line, the second pulmonary nodule feature vector is false negative;
all benign lung nodules are detected when the number of false positive second lung nodule feature vectors is minimum by optimizing mahalanobis space linear decision equation parameters.
Further optionally, the mahalanobis space linear decision equation satisfies the following formula:
wherein, the first and the second end of the pipe are connected with each other,
representing feature vectors x through jth benign feature subclass
The mahalanobis distance of the center of the circle,
representing feature vectors x through the ith malignancy feature subclass
The mahalanobis distance of the center of the circle,
is the ith malignancy feature subclass
The expansion times of the powder are increased by a plurality of times,
is a linear intercept parameter.
In a second aspect, the present invention also provides an apparatus for detecting benign and malignant lung nodules, comprising a processor, an input device, an output device, and a memory, the processor, the input device, the output device, and the memory being interconnected, the memory including the computer storage medium of the first aspect for detecting benign and malignant lung nodules, the processor being configured to invoke the program instructions. The device for detecting the benign and malignant pulmonary nodules provided by the invention has the advantages of compact structure and stable performance, and can efficiently and accurately execute the program instructions contained in the computer storage medium for detecting the benign and malignant pulmonary nodules in the memory.
Drawings
FIG. 1 is a schematic flow chart of program instructions stored on a computer storage medium for detecting benign and malignant pulmonary nodules according to the present invention;
FIG. 2 is a schematic diagram of an eight-domain pixel in an embodiment of the invention;
FIG. 3 is a schematic diagram of 4 different structural elements in an embodiment of the present invention;
FIG. 4 is a schematic diagram of a square field pixel in an embodiment of the invention;
FIG. 5 is a schematic diagram of a linear decision space in Mahalanobis space in an embodiment of the present invention;
fig. 6 is a schematic structural diagram of the device for detecting benign and malignant pulmonary nodules according to the present invention.
Detailed Description
Specific embodiments of the present invention will be described in detail below, and it should be noted that the embodiments described herein are merely illustrative and are not intended to limit the present invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, it will be apparent to one of ordinary skill in the art that: it is not necessary to employ these specific details to practice the present invention. In other instances, well-known circuits, software, or methods have not been described in detail in order to avoid obscuring the present invention.
Throughout the specification, reference to "one embodiment," "an embodiment," "one example" or "an example" means: the particular features, structures, or characteristics described in connection with the embodiment or example are included in at least one embodiment of the invention. Thus, the appearances of the phrases "in one embodiment," "in an embodiment," "one example" or "an example" in various places throughout this specification are not necessarily all referring to the same embodiment or example. Furthermore, the particular features, structures, or characteristics may be combined in any suitable combination and/or sub-combination in one or more embodiments or examples. Further, those of ordinary skill in the art will appreciate that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.
CT is considered as the best medical imaging means for examining lung diseases, but belongs to three-dimensional tomographic images, the amount of image data is large, especially HRCT (High Resolution computed tomography), the image data amount is rapidly increasing, and generally, one whole lung HRCT contains 200 to 500 scan images, or even more. The present invention is based on the insight that a large number of CT images of the lung assist a physician in intelligently identifying and diagnosing lung nodules in the early stages of lung cancer, and for its objects, features and advantages, it is further described in the following examples.
Referring to fig. 1, in one embodiment, the present invention provides a computer storage medium for detecting the malignancy and/or malignancy of a lung nodule, the computer storage medium storing a computer program comprising program instructions that, when executed by a processor, cause the processor to perform the steps of:
s01, acquiring lung electronic computed tomography images of known benign and malignant pulmonary nodules and pulmonary nodule benign and malignant pulmonary nodule conditions to be detected.
The pulmonary electron computed tomography images of the benign and malignant condition of a known lung nodule can be acquired by a hospital PACS (Picture Archiving and Communication System). The lung electron computer tomography image of the benign and malignant condition of the lung nodule to be detected can be acquired by a patient of the condition of the lung nodule to be detected by CT image acquisition in a hospital.
S02, lung nodule feature vectors in the lung electron computed tomography image are set.
In an alternative embodiment, the setting of the lung nodule feature vector in the lung electron computed tomography image in step S02 includes the following steps: setting two-dimensional feature vectors in a single-layer lung electron computed tomography image, wherein the two-dimensional feature vectors comprise two-dimensional morphological feature vectors, two-dimensional gray-scale feature vectors and two-dimensional gradient feature vectors; and superposing a plurality of single-layer lung electron computed tomography images to form a lung nodule three-dimensional image, and setting a three-dimensional characteristic vector of the lung nodule three-dimensional image, wherein the three-dimensional characteristic vector comprises a three-dimensional morphological characteristic vector, a three-dimensional gray characteristic vector and a three-dimensional ladder characteristic vector.
In this embodiment, for a two-dimensional morphological feature vector, a two-dimensional grayscale feature vector, and a two-dimensional gradient feature vector included in a two-dimensional feature vector, the two-dimensional grayscale feature vector and the two-dimensional gradient feature vector are further divided into an inner feature (inside), an expanded outer feature (outside), and a Separation feature for comparing the inside and the outside, respectively, according to a position feature.
Wherein the two-dimensional morphological feature vector comprises Size, area, circulation, eccentricity, compact, square compact, distance to center of current magnitude, distance to volume property, juxtapleural, fraction measuring magnitude, orientation X, orientation Y, X-Fraction, Y-Fraction, X-Fraction2 and Y-Fraction2. The Size represents the one-dimensional Size measurement of the lung nodule and is defined as the circular area with the diameter of the maximum distance between any two pixel points of the lung nodule boundary segmented from the lung electron computed tomography image; area represents the Area of a lung nodule region and is defined as the Area of each pixel point in a lung nodule image segmented from a lung electron computed tomography image multiplied by the number of the pixel points; circulation represents the roundness of the pulmonary nodule, and is defined as taking a circle with the area equal to the area of the pulmonary nodule area, taking the center of the circle to all pixel points in the pulmonary nodule area, calculating the number of the pixel points in the pulmonary nodule area in the circle, and dividing the maximum value of the number of the pixel points by the area of each pixel point; eccentricity represents Eccentricity, defined as the ratio of the maximum diameter of the lung nodule region perpendicular to the maximum diameter; compactness, defined as the area of the lung nodule region divided by the area of a circle of equal circumference; square Compactness represents Square Compactness and is defined as the area of a lung nodule region divided by the area surrounded by the minimum Square boundary of the lung nodule region; distance to Center of Current Lung represents the Euclidean Distance between the Center of the Lung nodule region and the Center of the bounding box of the Lung in the Current CT image where it is located; the Distance to lung perimeter represents the minimum value of the Euclidean Distance between the center of the lung nodule area and the half lung boundary pixel where the lung nodule area is located; juxtapleural indicates whether the lung nodule region is in contact with the lung boundary, juxtapleura is a Boolean feature; the Fraction touching lung nodule region represents the degree of contact of the lung nodule region with the lung boundary, defined as the number of pixels in contact with the lung boundary divided by the total number of its boundary pixels; orientation X represents the angle between the maximum diameter of the lung nodule area and the positive axis of X; origin Y represents the angle between the largest diameter of the lung nodule region and the Y forward axis; the X-Fraction is defined as taking the middle point of the upper boundary of a half lung boundary box where the lung nodule region is as a reference point, and dividing the horizontal distance between the central pixel point of the lung nodule region and the reference point by the horizontal width of the current boundary box; the Y-Fraction is defined as taking the middle point of the upper boundary of the half lung boundary box where the lung nodule area is as a reference point, and dividing the vertical distance between the central pixel point of the lung nodule area and the reference point by the vertical width of the current boundary box; defining X-Fraction2 as a reference point by taking the middle point of the boundary on the whole lung bounding box, and dividing the horizontal distance between the central pixel point of the lung nodule area and the reference point by the horizontal width of the current bounding box; Y-Fraction2 is defined as the vertical distance between the center pixel point of the lung nodule region and the reference point divided by the vertical width of the current bounding box, with the midpoint of the border over the entire lung bounding box as the reference point.
The two-dimensional gray feature vector is to calculate a statistical value of gray values of all pixel points in a two-dimensional region (inside) of the segmented lung nodule, and to calculate a statistical value of gray values of all pixel points in a peripheral region (outside) of the lung nodule through expansion operation aiming at the original segmented lung nodule region. Two-dimensional grayscale feature vectors include Max inside, min inside, mean inside, std definition inside, mean outside, std definition outside, contrast1, contrast2, mean separation, std definition separation, skiew inside, kurtosis inside, and motion 1 to 7 inside. The Max inside represents the maximum value of the gray value of each pixel point in the two-dimensional region of the lung nodule; min inside represents the minimum value of the gray value of each pixel point in the two-dimensional region of the lung nodule; mean inside represents the average value of the gray values of all pixel points in the two-dimensional region of the lung nodule; std definition inside represents the standard deviation of the gray value of each pixel point in the two-dimensional region of the lung nodule; mean outside represents the average value of gray values of all pixel points in the peripheral area of the lung nodule; std definition outside represents each pixel point in peripheral region of pulmonary noduleThe standard deviation of the gray value of (a); contrast1 is defined as
I.e. the mean value of the gray values of the pixels in the two-dimensional region of the lung nodule
Obtained by subtracting Mean outside of the peripheral region of the lung nodule; contrast2 is defined as
I.e. the standard deviation of the gray values of the pixels in the two-dimensional region of the lung nodule
Plus Std clearance outside of the peripheral region of the lung nodule; mean separation is defined as
Dividing the absolute value of the difference of the average values of the gray values of the pixels in the inner and outer regions of the lung nodule by the sum of the average values; std definition setting is defined as
I.e. the absolute value of the difference between the standard deviations of the gray values of the pixels in the inner and outer regions of the lung nodule is divided by the sum of the standard deviations; skew inside represents the statistical characteristic of asymmetric distribution of the pixels of the section of the lung nodule; kurtosis inside represents the statistical characteristic of the Kurtosis of the lung nodule section pixel; moment 1 to 7 inside represent 7 invariant moments of the lung nodule cross-section image, and the 7 invariant moments are firstly proposed by k.m. hu in 1962 to characterize the shape characteristics of the image. Statistical characterization of lung nodule cross-section pixels for the Skew inside and Kurtosis inside: assuming that the total number of N-level gray levels in the two-dimensional region of the lung nodule is N, and h (i) is used for representing the proportion of the number of the ith-level gray level pixels in the total number of the cross-section pixels, the average gray level value of the pixels in the two-dimensional region of the lung nodule is:
skaw insert satisfies the following formula:
kurtosis inside satisfies the following equation:
。
the characteristic amplitude (gradient magnitude), radial deviation (radial-deviation) and radial gradient (radial-gradient) of the two-dimensional planar gradient feature, and the interior (inside) of the two-dimensional region of the lung nodule, the inner boundary (perimeter) of the two-dimensional region of the lung nodule, the extra region (outer) obtained by morphological dilation operation on the lung nodule, and the outer boundary (perimeter) of the extra region are classified according to the position. The two-dimensional gradient feature vector includes: x gradient mean represents the average value of the X positive direction gradient amplitudes in the two-dimensional region of the lung nodule, Y gradient mean represents the average value of the Y positive direction gradient amplitudes in the two-dimensional region of the lung nodule, and XY gradient mean is defined as
The Gradient magnetic Mean instrument represents the average value of the two-dimensional Gradient amplitudes in the two-dimensional region of the lung nodule, the Gradient magnetic Std removal instrument represents the standard deviation of the two-dimensional Gradient amplitudes in the two-dimensional region of the lung nodule, the Gradient magnetic Mean permeator instrument represents the average value of the two-dimensional Gradient amplitudes in the inner boundary, the Gradient magnetic Std removal permeator instrument represents the standard deviation of the two-dimensional Gradient amplitudes in the inner boundary, the Gradient magnetic Mean outer represents the average value of the two-dimensional Gradient amplitudes in the additional region, the Gradient magnetic Std removal outer represents the standard deviation of the two-dimensional Gradient amplitudes in the additional region, the Gradient magnetic Mean outer represents the average value of the two-dimensional Gradient amplitudes in the additional region, the Gradient magnetic Mean outer represents the standard deviation of the two-dimensional Gradient amplitudes in the outer boundary, the Gradient magnetic Gradient amplitude is defined, the Gradient magnetic Mean outer boundary represents the standard deviation of the two-dimensional Gradient amplitudes in the additional region, and the Gradient outer boundary represents the standard deviation
Gradient magnetic Std removal Separation is defined as
Gradient magnetic Mean Perimeter Separation is defined as
The Gradient magnetic Std determination Perimeter Separation is defined as
Radial deviation Mean represents the Mean of two-dimensional Radial deviations within the two-dimensional region of the pulmonary nodule, radial deviation Std deviation Inside represents the standard deviation of two-dimensional Radial deviations within the two-dimensional region of the pulmonary nodule, radial deviation simulation inner represents the Mean of two-dimensional Radial deviations on the inner boundary, radial deviation Std deviation simulation inner represents the standard deviation of two-dimensional Radial deviations on the inner boundary, radial deviation Mean outer represents the Mean of two-dimensional Radial deviations in the additional region, radial deviation Std deviation outer represents the standard deviation of two-dimensional Radial deviations in the additional region, radial deviation Meeter represents the Mean of two-dimensional Radial deviations on the outer boundary, radial deviation Std deviation simulation outer represents the standard deviation of two-dimensional Radial deviations on the outer boundary, radial deviation is defined as the Mean of two-dimensional Radial deviations on the outer boundary
The Radial definition Std definition Separation is defined as
Radial definition Mean Perimeter Separation is defined as
A Radial definition Std definition Perimeter Separation is defined as
The Radial gradient Mean represents the average of the two-dimensional Radial gradients within the two-dimensional region of the lung nodule, the Radial gradient Mean represents the standard deviation of the two-dimensional Radial gradients within the two-dimensional region of the lung nodule, the Radial gradient Mean represents the average of the two-dimensional Radial gradients on the inner boundary, the Radial gradient Mean represents the standard deviation of the two-dimensional Radial gradients on the inner boundary, the Radial gradient Mean represents the average of the two-dimensional Radial gradients in the additional region, the Radial gradient Mean represents the standard deviation of the two-dimensional Radial gradients in the outer region, the Radial gradient Mean represents the average of the two-dimensional Radial gradients on the outer boundary, the Radial gradient Mean represents the standard deviation of the two-dimensional Radial gradients on the outer boundary, and the Radial gradient is defined as the average of the two-dimensional Radial gradients on the outer boundary
The Radial gradient Std removal Separation is defined as
Radial gradient Mean Perimeter Separation is defined as
And a Radial gradient Std depth Perimeter Separation as
. The two-dimensional plane gradient feature comprises 39 feature vectors, wherein the gradient feature of the two-dimensional plane is a two-dimensional vector which can be decomposed into two directions of X and Y for specific analysis (in the present specification, the X and Y directions are defined based on lung electron computed tomography images), that is, the two-dimensional plane gradient feature G can be expressed as
Ladder for any pixel point to be analyzedThe degree can be determined by first identifying the eight-domain pixels, and calculating with the first and last rows of the eight-domain pixels, as shown in fig. 2, the middle black pixel in fig. 2 is the center pixel, and the surrounding gray pixels are the eight-domain pixels, and the two-dimensional feature vector of the center pixel is determined
The components satisfy the following formula:
wherein, the first and the second end of the pipe are connected with each other,
expressing the gray value of the pixel point with the number i in the eight-domain pixels of the central pixel point,
representing the weight in the gradient calculation of the pixel point with the number i,
in which
，
Representing the coordinates of a central pixel
To the pixel point coordinate with number i
The euclidean distance between them. For two-dimensional feature vectors of central point pixels
The first and last rows of eight-domain pixels through which a component can pass are computed
The manner of the component is obtained. The feature amplitude (gradient magnitude) of the two-dimensional plane gradient feature G is defined as
(ii) a Radial-deviation (radial-deviation) is expressed as the included angle between the gradient vector on any central pixel point and the radial direction (the radial direction represents the connecting line of the central pixel point and the central point of the pulmonary nodule and points to the central direction of the pulmonary nodule)
(ii) a Radial-gradient (radial-gradient) is expressed as the projection size of the gradient vector of any central pixel point in the radial direction and is defined as
。
In the embodiment, the lung nodule three-dimensional image is formed by superposing a plurality of single-layer lung electron computed tomography images. Therefore, the three-dimensional feature vector comprises a two-dimensional statistical feature vector besides the three-dimensional feature vector of the three-dimensional feature vector, so that the comprehensiveness of three-dimensional feature extraction is ensured. The three-dimensional gray level feature vector and the three-dimensional gradient feature vector are further distinguished into an internal feature (inside), an expanded external feature (outside) and a Separation feature between the inside and the outside according to the position feature. The statistics of the two-dimensional feature values of the group of single-layer two-dimensional lung computer-scanned images constituting the three-dimensional image of the lung nodule represented by the two-dimensional statistical feature vector in the three-dimensional feature vector specifically include the following feature vectors:
wherein the content of the first and second substances,
representing the number of single-slice two-dimensional electronically computer-scanned images of the lungs that make up the three-dimensional image of the lung nodule,
representing the statistical average of the mean values of the two-dimensional characteristic values of the single-layer two-dimensional lung computer-scanned images constituting the three-dimensional image of the lung nodule,
representing the statistical minimum of the two-dimensional characteristic values of the single-layer two-dimensional lung computer scanning image which forms the three-dimensional image of the lung nodule,
representing the statistical maximum of the two-dimensional characteristic values of the single-layer two-dimensional lung electronic computer scanning image which forms the three-dimensional image of the lung nodule,
and (3) representing the statistical standard deviation of the two-dimensional characteristic values of the single-layer two-dimensional lung electronic computer scanning image forming the three-dimensional image of the lung nodule.
The three-dimensional morphological feature vector in the three-dimensional feature vector of the lung nodule three-dimensional image is mainly obtained by extending two-dimensional morphological features, and specifically comprises the following steps: size, volume, area Difference, uniformity, integration, compact 3D, cube compact, project compact, distance to center of current Volume 3D, distance to center of project Volume, distance to Volume surface, juxtaplual 3D, fraction creating Volume 3D, organization X3D, organization Y3D, organization Z3D, X-Fraction 3D, Y-Fraction 3D, Z-Fraction 3D, X-Fraction 2D, Y-Fraction 2D, Z-Fraction 2D. The Size represents the maximum value of the distance between any two pixel points on the surface of the three-dimensional lung nodule, namely the diameter of an external sphere; volume represents the Volume of the three-dimensional lung nodule, and can be obtained by counting the number of three-dimensional pixels of the three-dimensional lung nodule and multiplying the number by the Volume of each three-dimensional pixel; the Area Difference represents the average variation of the areas of all the divided two-dimensional regions of a lung nodule, namely the sum of the variations between every two adjacent segments is divided by the number of the variations; the Sphericity of the three-dimensional pulmonary nodules is expressed by the Sphericity, the Sphericity is defined as taking a sphere with the same volume as the three-dimensional pulmonary nodules, the sphere center is taken through all pixel points in a pulmonary nodule body, meanwhile, the pixel number of the three-dimensional pulmonary nodules falling into the sphere is calculated, the maximum value of the number is taken to be multiplied by the volume of each pixel, and then the maximum value is divided by the volume of the sphere to be taken as the characteristic value of the Sphericity; elongation represents the eccentricity in three dimensions, defined as the ratio of the maximum diameter perpendicular to the maximum diameter of the three-dimensional pulmonary nodule to the maximum diameter; compact 3D represents sphere Compactness, defined as the volume of a three-dimensional lung nodule divided by the volume of a sphere of equal surface area; cube compact is defined as the ratio of the volume of a three-dimensional lung nodule to the volume of its bounding box in three-dimensional space; project compatibility is located as the ratio of the projected area of the three-dimensional lung nodule projected onto the x-y plane to the projected bounding box area; distance to center of current lung 3D is defined as the Euclidean Distance between the center of the three-dimensional lung nodule and the center of the whole lung. Wherein the center of the lung is the center of its bounding box; the center of the lung nodule is determined by a sphere which has the same volume with the lung nodule, the sphere and the lung nodule are superposed in a three-dimensional space, and when the superposed volume is the maximum, the position of the center of the sphere is the center of the lung nodule; distance to center of projected lung is defined as the Euclidean Distance between the center of the three-dimensional lung nodule and the center of the whole lung projected onto the x-y plane; distance to lung surface 3D is expressed as the minimum of the euclidean Distance between the center pixel of the three-dimensional lung nodule and the lung surface pixel; juxtapleural 3D represents whether the three-dimensional lung nodule region is in contact with a three-dimensional lung boundary, juxtapleura 3D is a Boolean feature, fraction touchhing 3D represents the contact degree of the three-dimensional lung nodule region and the three-dimensional lung boundary, and Fraction touchhing 3D is a Boolean feature; orientation X3D represents the included angle between the maximum diameter of a three-dimensional lung nodule in a three-dimensional space and the X axis; orientation Y3D represents the included angle between the maximum diameter of the three-dimensional pulmonary nodule in a three-dimensional space and the Y axis; orientation Z3D represents the included angle between the maximum diameter of the three-dimensional pulmonary nodule in a three-dimensional space and the Z axis; the X-Fraction 3D is defined as that the midpoint of the boundary in the X-axis direction of the boundary box of the half lung where the three-dimensional lung nodule is located is used as a reference point, and the distance between the center of the three-dimensional lung nodule and the reference point in the direction opposite to the X-axis direction is divided by the width of the boundary in the X-axis direction of the current boundary box; the Y-Fraction 3D is defined as taking the middle point of the boundary box of the half lung where the three-dimensional lung nodule is located in the Y-axis direction as a reference point, and dividing the distance between the center of the three-dimensional lung nodule and the reference point in the relative Y-axis direction by the width of the boundary box in the Y-axis direction at present; the Z-Fraction 3D is defined as taking the middle point of the boundary box of the half lung where the three-dimensional lung nodule is located in the Z-axis direction as a reference point, and dividing the distance between the center of the three-dimensional lung nodule and the reference point in the relative Z-axis direction by the width of the boundary box in the Z-axis direction; the X-Fraction 2D is defined as that the midpoint of the boundary in the X-axis direction of the whole lung boundary box is taken as a reference point, and the distance between the center of the three-dimensional lung nodule and the reference point in the relative X-axis direction is divided by the boundary width in the X-axis direction of the current boundary box; the Y-Fraction 2D is defined as that the midpoint of the boundary in the Y-axis direction of the boundary box of the whole lung is taken as a reference point, and the distance between the center of the three-dimensional lung nodule and the reference point in the relative Y-axis direction is divided by the boundary width in the Y-axis direction of the current boundary box; the Z-Fraction 2D is defined as taking the middle point of the boundary box of the whole lung in the Z-axis direction as a reference point, and dividing the distance between the center of the three-dimensional lung nodule and the reference point in the Z-axis direction by the width of the boundary box in the Z-axis direction.
The three-dimensional gray level feature vector in the three-dimensional feature vector is a three-dimensional mapThe statistics of the gray level in the image are specifically divided into the following three aspects: statistical aspects (features of the statistical aspects include Maximum, minimum, mean, std, standard deviation, skew, kurtosis, etc.)), regional aspects (features of the regional aspects include, inner, outer 1, outer 2, outer 3, outer Z, above, and below), and comparative aspects (features of the comparative aspects include, separation1, separation2, separation3, separation Z, constrast 11, constrast 12, constrast 13, constrast 1Z, constrast 21, constrast 22, constrast 23, and constrast 2Z). Each specific three-dimensional gray level feature vector is formed by combining features in a statistical aspect and features in a region aspect or features in a contrast aspect, for example: and a three-dimensional gray feature vector such as Maximum _ inside and Maximum _ section 1. Regional characteristics are based on the division of different regions, in this embodiment, inside represents the internal region of the three-dimensional lung nodule, please refer to fig. 3, and the original three-dimensional lung nodule is obtained through the dilation operation in mathematical morphology by different 4 structural elements: the structural element a is a spheroid structural element, and in an x-y plane corresponding to a three-dimensional pulmonary nodule, by taking a pixel layer of a circle with the radius of 5 pixel points as a center, the same pixel layer is respectively expanded up and down in the z-axis direction to obtain an outside1 region of the spheroid structural element, as shown in fig. 3 (a); the structural element b is also a spheroid structural element, and in an x-y plane corresponding to the three-dimensional pulmonary nodule, by taking a pixel layer of a circle with a radius of 10 pixel points as a center, the same pixel layer is respectively expanded up and down in the z-axis direction to obtain outside2 regions of the spheroid structural element, as shown in fig. 3 (b); the structural element c is a circular structural element, that is, a circular pixel layer with radius of x-y plane corresponding to the three-dimensional pulmonary nodule being 10 pixel points obtains outside3 region of the circular structural element, as shown in fig. 3 (c); the structural element d is a columnar structural element, one pixel of the x-y plane corresponding to the three-dimensional lung nodule serves as a pixel layer, and then such a layer of pixels is expanded up and down in the z-axis direction to obtain an outsolid z region, as shown in fig. 3 (d). Above indicates the maximum value of the three-dimensional pulmonary nodule in the Z axisThe area obtained by projecting the upper layer of one pixel point on the x-y plane. Below represents the area obtained by the projection of the next layer of the three-dimensional pulmonary nodule on the x-y plane at the pixel point with the minimum Z-axis value. The features for the comparison aspect, namely, segmentation 1, segmentation 2, segmentation 3, segmentation Z, concentrate 11, concentrate 12, concentrate 13, concentrate 1Z, concentrate 21, concentrate 22, concentrate 23 and concentrate 2Z, can be classified into three categories: seperationi, concentrate 1I, and concentrate 2I. Wherein, section I represents the difference between a certain characteristic of an inside region and the same characteristic vector in outsid I (I =1,2,3, Z), and the section I specifically satisfies the following formula:
(ii) a The contist 1I represents a class of characteristic value of the region minus the corresponding class of characteristic value of the outseI (I =1,2,3, Z) in the inside region, and the contist 1I specifically satisfies the following formula
(ii) a Conttrast 2I represents the ratio of conttrast 1I divided by the sum of the standard deviations of the features in the inside and outeideI regions, and is specifically defined as
And are and
indicate the standard deviation corresponding to the same class of features in the inner and outer idei regions, respectively.
The three-dimensional gradient features include: three-dimensional gradient feature vector
Component features in X, Y, Z axes
、
、
Three-dimensional gradient feature vector
Contrast feature vector and three-dimensional gradient feature vector on X and Y axes
Contrast feature vector and three-dimensional gradient feature vector on X and Z axes
Contrast feature vector and three-dimensional gradient feature vector on Z axis and Y axis
Gradient magnitude feature (gradient magnitude) of (1), three-dimensional gradient feature vector
Radial-deviation characterization, three-dimensional gradient eigenvectors
Radial-gradient feature (radial-gradient). The method comprises the steps of obtaining a lung nodule expansion operation, wherein the lung nodule expansion operation comprises a gradient amplitude characteristic (gradient magnitude), a radial deviation characteristic (radial-deviation) and a radial gradient characteristic (radial-gradient), and further comprises the steps of classifying the interior (inside) of a two-dimensional region of the lung nodule, the inner boundary (perimeter inside) of the two-dimensional region of the lung nodule, an extra region (outeidel, I =1,2,3, Z) obtained after the lung nodule expansion operation by morphology, and the outer boundary (perimeter outeidel) of the extra region according to positions, wherein a specific classification mode and a specific calculation mode can be obtained by analogy through two-dimensional plane gradient characteristic operation. Three-dimensional gradient feature vector
The three-dimensional vector can be decomposed into three-dimensional analysis in X, Y and Z directions (in the specification, the X, Y and Z directions are three-dimensional analysis formed based on single-layer lung electron computer tomography image definitionImage defined), i.e. three-dimensional gradient feature vectors
Can be expressed as
For the gradient of any pixel point to be analyzed, as shown in fig. 4, a square field pixel can be confirmed first, 9 pixels on the upper layer are numbered 1 to 9 sequentially from top to bottom in the square field pixel from a center point pixel to the left, and 9 pixels on the lower layer are numbered 10 to 18 sequentially in the same order, wherein the components of the three-dimensional feature vector are numbered
The components satisfy the following formula:
wherein, the first and the second end of the pipe are connected with each other,
representing the gray value of the pixel point with the number i in the square field pixel,
the weight of the pixel point numbered i is represented,
which is
In the step (1), the first step,
representing the coordinates of a central pixel point
To the pixel point coordinate with number i
The euclidean distance between them. Component of three-dimensional feature vector for center point pixel
And component(s)
Can be obtained in the same manner. Three dimensional gradient feature
Is defined as a characteristic magnitude of
(ii) a Radial-deviation (radial-deviation) is expressed as the included angle between the gradient vector on any central pixel point and the radial direction (the radial direction represents the connecting line of the central pixel point and the center point of the pulmonary nodule and points to the central direction of the pulmonary nodule)
(ii) a Radial-gradient (radial-gradient) is expressed as the projection size of the gradient vector of any central pixel point in the radial direction and is defined as
。
In the step S02, a plurality of characteristic vectors for representing the characteristics of the lung nodules are set according to the two-dimensional and three-dimensional characteristics of the lung nodules in the lung electron computed tomography image, so that comprehensive and accurate theory and data support are provided for subsequent extraction and judgment of the lung nodules to be judged, and the accuracy of judgment results is improved.
S03, extracting a first pulmonary nodule feature vector in a pulmonary electronic computed tomography image of the known good and malignant condition of the pulmonary nodule.
The extracting of the first lung nodule feature vector in the lung CT image of the known good and malignant lung nodule in step S03 may be performed by writing a corresponding program to perform corresponding extraction on the feature vector in the lung CT image of the known good and malignant lung nodule according to the basic definition of the feature vector in step S02. And then, calculating the characteristic value through the extracted characteristic vector, thereby obtaining all information of the first lung nodule characteristic vector, namely the first lung nodule characteristic vector with the corresponding characteristic value.
And S04, obtaining various clustering results through the first lung nodule characteristic vector, and generating a reference clustering data set through the various clustering results.
In step S04, obtaining a plurality of clustering results through the first lung nodule feature vector, and generating a reference clustering data set through the plurality of clustering results, including the following steps: carrying out clustering analysis on the first lung nodule characteristic vector by utilizing different clustering algorithms to obtain a plurality of clustering results, or carrying out clustering analysis on the first lung nodule characteristic vector by utilizing the same clustering algorithm with different parameters and different initial values to obtain a plurality of clustering results; and combining the multiple clustering results to obtain a combined result, and carrying out clustering analysis on the combined result to obtain a reference clustering data set.
Hierarchical clustering is a static clustering algorithm, which is divided into a merging algorithm and a splitting algorithm. The merging algorithm reduces the number of cluster centers at each step, and the clustering results from the merging of the two clusters from the previous step. The splitting algorithm is opposite to the merging algorithm in principle, the number of the clustering centers is increased in each step, and the result generated by clustering in each step is obtained by splitting one clustering center in the previous step into two clustering centers. The Kmeans clustering method is a dynamic clustering algorithm, and can minimize the sum of squares of distances from all samples in a clustering domain to a clustering center. The clustering method for obtaining multiple clustering results in step S04 includes, but is not limited to: and clustering algorithms such as a Kmeans clustering method, an HC clustering algorithm, a MeanShift algorithm, a Kmeans & HC combined clustering algorithm and the like. According to the method, the first lung nodule characteristic vector is subjected to multiple clustering analysis through a dynamic clustering algorithm and a static clustering algorithm, and multiple clustering results obtained are strong in robustness. The combination of the multiple clustering results to obtain the combined result includes but is not limited to a Kmeans clustering method, an HC clustering algorithm, a MeanShift algorithm, a Kmeans & HC combined clustering algorithm and other clustering algorithms.
In an alternative embodiment, a Kmeans clustering method is adopted, and a clustering combination of multiple clustering results is obtained by performing clustering analysis on the first lung nodule feature vector by using different k values and initial clustering centers: p = { P 1 ,P 2 ,…, P N Within a certain range of k min ,k max ]Inner variation, each time clustering selects a different initial cluster center. Each P l Obtaining an autocorrelation matrix C l Autocorrelation matrix C l The following formula is satisfied:
combining a plurality of clustering results in the clustering combination through a voting mechanism to obtain a combined result as follows:
wherein the content of the first and second substances,
a matrix obtained by combining a plurality of kinds of clustering results is represented,
，
representation matrix
Middle sample
And samples
The corresponding value at the indicated position is,
representing the total number of clustering results. Then the combined result is hierarchically clustered, and a single connection method SL is selected to connect the matrix
As a sample distance matrix for the SL algorithm. Defining a threshold value to determine the final clustering result of SL, wherein the threshold value is the lifetime of k-cluster division and is marked as l k The lifetime is the difference in ordinate from the level at which the k classes are generated to the level at which the k-1 clusters are generated. And finally, taking the layer with the largest lifetime as a final clustering layer, and obtaining a final clustering result which is the reference clustering data set. The feature vectors in the reference cluster dataset have already calibrated the benign and malignant properties of the corresponding lung nodule.
S05, extracting a second pulmonary nodule feature vector in the lung electronic computed tomography image of the benign and malignant condition of the pulmonary nodule to be detected.
The method for extracting the second lung nodule feature vector in the lung electronic computed tomography image of the benign and malignant condition of the lung nodule in the step S05 can be similar to the method for extracting the first lung nodule feature vector in the step S03.
And S06, detecting benign lung nodules and malignant lung nodules in the lung electron computed tomography image of the benign and malignant condition of the lung nodules to be detected by using the second lung nodule feature vector and combining the reference clustering data set.
Step S06, detecting benign lung nodules and malignant lung nodules in the lung electron computed tomography image of the benign and malignant condition of the lung nodules to be detected by using the second lung nodule feature vector in combination with the reference cluster data set, including the following steps:
s601, obtaining lung nodule feature vector x aiming at ith malignancy feature subclass in reference clustering data set respectively
Bayes posterior probability of (2)
And the jth benign feature subclass
Bayes posterior probability of
. Bayes posterior probability
And
respectively satisfy the following formulas:
，
wherein, in the step (A),
representing a malignancy feature subclass
The average of the feature vectors of the respective samples,
representing a malignant feature subclass
The covariance matrix of the eigenvectors of each sample,
indicating to subclass malignancy characteristics
Distribution of
The number of times of the total number of the components,
representing a malignant feature subclass
The average of the feature vectors of the respective samples,
representing a malignancy feature subclass
The covariance matrix of the eigenvectors of each sample,
representing benign feature subclasses
The average of the feature vectors of the respective samples,
benign feature subclass
The covariance matrix of the eigenvectors of each sample.
S602, setting a quality determination boundary
The benign-malignant determination boundary satisfies the following equation:
。
and S603, converting the boundary formula into a Markov space linear decision equation through a good and malignant decision boundary formula. And (3) solving the natural logarithm of two sides of the good and malignant judgment boundary formula, and obtaining the following formula after deformation:
order to
It can be further simplified to the following formula:
wherein, the first and the second end of the pipe are connected with each other,
representing feature vectors x through jth benign feature subclass
The mahalanobis distance of the center of the circle,
representing feature vectors x through the ith malignancy feature subclass
The mahalanobis distance of the center of the circle,
is the ith malignancy feature subclass
The expansion times are increased by a plurality of times,
is a linear intercept parameter.
S604, obtaining the Bayes posterior probability of the second pulmonary nodule feature vector
Value of (d) and Bayesian posterior probability
The value of (c).
S605, preliminarily judging the benign and malignant degree of the second pulmonary nodule characteristic vector according to a benign and malignant degree judgment boundary formula, wherein the benign and malignant degree judgment satisfies the following formula:
s606, respectively calculating the second lung nodule feature vector to the jth benign feature subclass
Mahalanobis distance of center
And to the ith malignancy feature subclass
Mahalanobis distance of center
。
S607, according to the Mahalanobis distance
And said Mahalanobis distance
Combined decision coordinates
At the mahalanobis location, the second lung nodule feature vector is again determined to be benign or malignant:
if the second lung nodule feature vector belongs to the benign feature after the primary judgment, the judgment coordinate corresponding to the second lung nodule feature vector
If the second lung nodule characteristic vector is positioned below the Mahalanobis space linear judgment straight line, the second lung nodule characteristic vector is false positive; if the second lung nodule feature vector is judged to belong to the malignant feature for the first time, judging coordinates corresponding to the second lung nodule feature vector
And if the second lung nodule feature vector is positioned above the linear judgment straight line in the Mahalanobis space, the second lung nodule feature vector is false negative.
And S608, optimizing the Mahalanobis space linear decision equation parameters to detect all benign lung nodules when the number of the second lung nodule feature vectors which are false positive is minimum. See FIG. 5 for a view in Mahalanobis spaceIn the linear judgment space, the abscissa is the mahalanobis distance of the distribution center of a certain subclass of the second pulmonary nodule feature vector and the malignant classification in the feature space, the ordinate is the mahalanobis distance equation of the distribution center of a certain subclass of the second pulmonary nodule feature vector and the benign classification in the feature space, and when the parameters in the mahalanobis space linear judgment equation are adjusted
When the value is taken, the change of the straight line in the mahalanobis space linearity determination space is as shown in fig. 5 (a); when adjusting parameters in the mahalanobis space linear decision equation
When the values are taken, the change of the straight line in the mahalanobis space linearity determination space is as shown in fig. 5 (b). By varying parameters
And
and an optimal linear discriminant function can be obtained, so that all negative lung nodules can be detected when the number of false-positive second lung nodule feature vectors is minimum, and the accuracy of judgment is greatly guaranteed.
The invention extracts a plurality of characteristic vectors of lung electron computer tomography images of known benign and malignant conditions of lung nodules, generates a reference cluster data set for judging the benign and malignant conditions of the lung nodules after cluster analysis, and identifies the extracted second lung nodule characteristic vectors and diagnoses the benign and malignant conditions of the lung nodules to be judged by using the reference cluster data set. According to the method, a stable and reliable lung nodule reference clustering data set is generated through clustering analysis of a large number of lung nodule benign and malignant data judged by medical clinical certification and doctor experience, efficient and rapid identification and diagnosis of the lung nodule to be judged are realized through the reference clustering data set, and doctors are effectively assisted in diagnosing and treating benign and malignant pathologies of the lung nodule.
Referring to fig. 6, the present invention also provides an apparatus for detecting benign and malignant pulmonary nodules comprising one or more processors 601; one or more input devices 602, one or more output devices 603, and memory 604. The processor 601, the input device 602, the output device 603, and the memory 604 are connected by a bus 605. The memory 604 may be the computer readable storage medium of any one of the aspects provided herein, and the processor 601 is configured to invoke the program instructions contained in the computer storage medium provided herein for detecting benign and malignant lung nodules.
In an alternative embodiment, processor 601 may be a Central Processing Unit (CPU), which may be other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an FPGA (Field-Programmable Gate Array) or other Programmable logic device, discrete Gate or transistor logic, discrete hardware components, or the like. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The input device 602 may be used to acquire an electron computed tomography image of the lung of a known good-malignancy condition of a lung nodule and a condition of a lung nodule to be detected. The output device 603 may be configured to output the benign and malignant condition of the lung nodule to be detected, which is obtained through program instructions stored in any one of the computer-readable storage media provided in the present invention, to a target terminal for displaying. The memory 604 may include both read-only memory and random access memory, and provides instructions and data to the processor 601. A portion of the memory 604 may also include non-volatile random access memory. For example, the memory 604 may also store device type information.
In yet another alternative embodiment, the processor 601, the input device 602, and the output device 603 described in this embodiment of the present invention may execute an implementation manner described in a program instruction stored in any computer-readable storage medium provided by the present invention, and may also execute an implementation manner of a terminal device described in this embodiment of the present invention, which is not described herein again. The computer-readable storage medium may be an internal storage unit of the terminal device in any of the foregoing embodiments, for example, a hard disk or a memory of the terminal device. The computer-readable storage medium may also be an external storage device of the terminal device, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), and the like provided on the terminal device. Further, the computer-readable storage medium may include both an internal storage unit and an external storage device of the terminal device. The computer-readable storage medium is used for storing the computer program and other programs and data required by the terminal device. The above-described computer-readable storage medium may also be used to temporarily store data that has been output or is to be output.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; such modifications and substitutions do not depart from the spirit and scope of the present invention, and they should be construed as being included in the following claims and description.
Claims (10)
1. A computer storage medium for detecting pulmonary nodule malignancy and well, the computer storage medium storing a computer program comprising program instructions that, when executed by a processor, cause the processor to perform the steps of:
acquiring lung electron computer tomography images of known good and malignant conditions of lung nodules and good and malignant conditions of lung nodules to be detected;
setting a lung nodule characteristic vector in a lung electron computed tomography image;
extracting a first pulmonary nodule feature vector in a pulmonary electron computed tomography image of a known benign and malignant condition of a pulmonary nodule;
obtaining a plurality of clustering results through the first lung nodule feature vector, and generating a reference clustering data set through the plurality of clustering results;
extracting a second pulmonary nodule characteristic vector in the lung electronic computed tomography image of the benign and malignant condition of the pulmonary nodule to be detected;
and detecting benign lung nodules and malignant lung nodules in the lung electron computed tomography image of the benign and malignant condition of the lung nodules to be detected by utilizing the second lung nodule feature vector and combining the reference cluster data set.
2. The computer storage medium for detecting benign and malignant lung nodules according to claim 1, wherein said setting lung nodule feature vectors in lung electron computed tomography images comprises the steps of:
setting two-dimensional feature vectors in a single-layer lung electron computed tomography image, wherein the two-dimensional feature vectors comprise two-dimensional morphological feature vectors, two-dimensional gray-scale feature vectors and two-dimensional gradient feature vectors;
and superposing a plurality of single-layer lung electron computed tomography images to form a lung nodule three-dimensional image, and setting a three-dimensional characteristic vector of the lung nodule three-dimensional image, wherein the three-dimensional characteristic vector comprises a three-dimensional morphological characteristic vector, a three-dimensional gray characteristic vector and a three-dimensional ladder characteristic vector.
3. The computer storage medium for detecting benign and malignant lung nodules according to claim 2, wherein said three-dimensional grayscale feature vector comprises a contrast-wise grayscale feature vector comprising:
wherein, inside represents the inner region of the three-dimensional lung nodule,
，
representing the expanded peripheral region of the three-dimensional lung nodule,
and
are shown in imide and
the same class of features within a region may,
and
are shown in imide and
and standard deviations corresponding to the same type of features in the region respectively.
4. The computer storage medium for detecting benign and malignant lung nodules according to claim 2, wherein said three-dimensional gradient feature vectors comprise components of gradient feature vectors on three-dimensional coordinate axes
Component (a) of
The following formula is satisfied:
wherein the content of the first and second substances,
，
representing the components of the three-dimensional gradient feature vector on the X-axis,
representing the components of the three-dimensional gradient feature vector in the Y-axis,
representing the component of the three-dimensional gradient feature vector in the Z-axis,
the gray value of a pixel point which is numbered as i in pixels in the square field of a central pixel point of the three-dimensional ladder characteristic vector is represented,
representing the weight of the pixel point numbered i,
in which
，
，
Representing the coordinates of a central pixel
To the pixel point coordinate with number i
The euclidean distance between them.
5. The computer storage medium for detecting benign and malignant lung nodules according to claim 1, wherein said obtaining a plurality of kinds of clustering results through said first lung nodule feature vector, and generating a reference clustering dataset through said plurality of kinds of clustering results, comprises the steps of:
carrying out clustering analysis on the first lung nodule characteristic vector by utilizing different clustering algorithms to obtain a plurality of clustering results; or alternatively
Performing clustering analysis on the first pulmonary nodule characteristic vector through the same clustering algorithm with different parameters and different initial values to obtain a plurality of clustering results;
and combining the multiple clustering results to obtain a combined result, and performing clustering analysis on the combined result to obtain a reference clustering data set.
6. The computer storage medium for detecting benign and malignant pulmonary nodules according to claim 5, wherein said different clustering algorithms comprise a Kmeans clustering method, a HC clustering algorithm, a MeanShift algorithm, and a Kmeans-HC combination clustering algorithm.
7. The computer storage medium for detecting benign and malignant lung nodules according to claim 5, wherein said combined result satisfies the following formula:
wherein, the first and the second end of the pipe are connected with each other,
a matrix obtained by combining a plurality of kinds of clustering results is represented,
，
representation matrix
Middle sample
And samples
The corresponding value at the indicated position is,
，
representing the total number of clustering results.
8. The computer storage medium for detecting benign and malignant lung nodules according to claim 1, wherein said detecting benign and malignant lung nodules in the lung EMCT images of the condition of benign and malignant lung nodules to be detected using said second lung nodule feature vector in combination with said reference cluster data set comprises the steps of:
obtaining pulmonary nodule feature vectors
Pin for sortingFor the ith malignancy feature subclass in the reference cluster dataset
Bayes posterior probability of (2)
And the jth benign feature subclass
Bayes posterior probability of
；
Setting a quality/malignancy determination boundary
The benign-malignant determination boundary satisfies the following formula:
converting the good and malignant judgment boundary formula into a Markov space linear judgment equation;
obtaining a Bayesian posterior probability of the second pulmonary nodule feature vector
And Bayes posterior probability
；
And preliminarily judging the benign and malignant degree of the second lung nodule characteristic vector according to a benign and malignant degree judgment boundary formula, wherein the benign and malignant degree judgment satisfies the following formula:
respectively calculating the second lung nodule feature vector to the jth benign feature subclass
Mahalanobis distance of center
And to the ith malignancy feature subclass
Mahalanobis distance of center
；
According to the Mahalanobis distance
And said Mahalanobis distance
Combined decision coordinates
At the mahalanobis location, the second lung nodule feature vector is again determined to be benign or malignant: if the second lung nodule feature vector belongs to the benign feature after the primary judgment, the judgment coordinate corresponding to the second lung nodule feature vector
If the second lung nodule feature vector is located below the mahalanobis space linear judgment straight line, the second lung nodule feature vector is false positive; if the second lung nodule feature vector is judged to belong to the malignant feature for the first time, judging coordinates corresponding to the second lung nodule feature vector
Is located above the linear judgment straight line of the Mahalanobis space, thenThe second lung nodule feature vector is false negative;
all benign lung nodules are detected by optimizing the mahalanobis space linear decision equation parameters such that the number of false positive second lung nodule feature vectors is minimized.
9. The computer storage medium for detecting benign and malignant lung nodules according to claim 8, wherein said mahalanobis space linear decision equation satisfies the following formula:
wherein the content of the first and second substances,
representing feature vectors x through the jth benign feature subclass
The mahalanobis distance of the center of the circle,
representing feature vectors x through the ith malignancy feature subclass
The mahalanobis distance of the center of the circle,
is the first malignancy feature subclass
The expansion times are increased by a plurality of times,
is a linear intercept parameter.
10. An apparatus for detecting the malignancy and/or malignancy of a lung nodule, comprising a processor, an input device, an output device, and a memory interconnected with one another, the memory including a computer storage medium according to any one of claims 1 to 9, the processor being configured to invoke the program instructions.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211552686.6A CN115661132A (en) | 2022-12-05 | 2022-12-05 | Computer storage medium and device for detecting benign and malignant pulmonary nodules |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211552686.6A CN115661132A (en) | 2022-12-05 | 2022-12-05 | Computer storage medium and device for detecting benign and malignant pulmonary nodules |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115661132A true CN115661132A (en) | 2023-01-31 |
Family
ID=85024603
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211552686.6A Pending CN115661132A (en) | 2022-12-05 | 2022-12-05 | Computer storage medium and device for detecting benign and malignant pulmonary nodules |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115661132A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116485778A (en) * | 2023-04-27 | 2023-07-25 | 鄂东医疗集团市中心医院 | Imaging detection method, imaging detection system, computer equipment and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110142301A1 (en) * | 2006-09-22 | 2011-06-16 | Koninklijke Philips Electronics N. V. | Advanced computer-aided diagnosis of lung nodules |
| CN103745227A (en) * | 2013-12-31 | 2014-04-23 | 沈阳航空航天大学 | Method for identifying benign and malignant lung nodules based on multi-dimensional information |
| CN109036547A (en) * | 2018-06-11 | 2018-12-18 | 燕山大学 | A kind of lung CT image computer aided system and method based on clustering |
| CN111008962A (en) * | 2019-11-27 | 2020-04-14 | 哈尔滨理工大学 | Chest CT pulmonary nodule automatic check out system |
| CN112819747A (en) * | 2019-10-31 | 2021-05-18 | 深圳市华嘉生物智能科技有限公司 | Method for automatically diagnosing benign and malignant nodules based on lung tomography image |
-
2022
- 2022-12-05 CN CN202211552686.6A patent/CN115661132A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110142301A1 (en) * | 2006-09-22 | 2011-06-16 | Koninklijke Philips Electronics N. V. | Advanced computer-aided diagnosis of lung nodules |
| CN103745227A (en) * | 2013-12-31 | 2014-04-23 | 沈阳航空航天大学 | Method for identifying benign and malignant lung nodules based on multi-dimensional information |
| CN109036547A (en) * | 2018-06-11 | 2018-12-18 | 燕山大学 | A kind of lung CT image computer aided system and method based on clustering |
| CN112819747A (en) * | 2019-10-31 | 2021-05-18 | 深圳市华嘉生物智能科技有限公司 | Method for automatically diagnosing benign and malignant nodules based on lung tomography image |
| CN111008962A (en) * | 2019-11-27 | 2020-04-14 | 哈尔滨理工大学 | Chest CT pulmonary nodule automatic check out system |
Non-Patent Citations (4)
| Title |
|---|
| KISELEV ET AL.: "SC3 - consensus clustering of single-cell RNA-Seq data", 《EUROPE PMC FUNDERS GROUP》, vol. 14, no. 5, pages 483 - 486 * |
| TE BRAKE G M ET AL.: "An automatic method to discriminate malignant masses from normal tissue in digital mammograms∗", 《PHYSICS IN MEDICINE & BIOLOGY》, vol. 45, no. 10, pages 2843 - 2857 * |
| 吴颜艳: "基于多组学数据和机器学习的肾癌致病因素研究", 《中国优秀硕士学位论文全文数据库 医药卫生科技辑》, no. 1, pages 30 - 31 * |
| 周帅: "基于CT影像的肺结节辅助诊断系统的设计与实现", 《中国优秀硕士学位论文全文数据库 医药卫生科技辑》, no. 4, pages 6 - 20 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116485778A (en) * | 2023-04-27 | 2023-07-25 | 鄂东医疗集团市中心医院 | Imaging detection method, imaging detection system, computer equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | A hybrid CNN feature model for pulmonary nodule malignancy risk differentiation | |
| US7346209B2 (en) | Three-dimensional pattern recognition method to detect shapes in medical images | |
| JP5864542B2 (en) | Image data processing method, system, and program for detecting image abnormality | |
| Ye et al. | Shape-based computer-aided detection of lung nodules in thoracic CT images | |
| Campadelli et al. | A fully automated method for lung nodule detection from postero-anterior chest radiographs | |
| El-Baz et al. | Three-dimensional shape analysis using spherical harmonics for early assessment of detected lung nodules | |
| Shaukat et al. | Computer-aided detection of lung nodules: a review | |
| Liu et al. | A CADe system for nodule detection in thoracic CT images based on artificial neural network | |
| US9014456B2 (en) | Computer aided diagnostic system incorporating appearance analysis for diagnosing malignant lung nodules | |
| JP4660546B2 (en) | Method for characterizing objects in digitized images and computer-readable program storage | |
| Hasan et al. | Automated screening of MRI brain scanning using grey level statistics | |
| Kim et al. | The recent progress in quantitative medical image analysis for computer aided diagnosis systems | |
| JP2018175217A (en) | Image processing apparatus, method and program | |
| EP4118617A1 (en) | Automated detection of tumors based on image processing | |
| Li et al. | Segmentation of pulmonary nodules using a GMM fuzzy C-means algorithm | |
| Jaffar et al. | Ensemble classification of pulmonary nodules using gradient intensity feature descriptor and differential evolution | |
| CN115661132A (en) | Computer storage medium and device for detecting benign and malignant pulmonary nodules | |
| WO2022225794A1 (en) | Systems and methods for detecting and characterizing covid-19 | |
| JP6987721B2 (en) | Image processing equipment, methods and programs | |
| Farag et al. | Parametric and non-parametric nodule models: Design and evaluation | |
| Altun et al. | Feasibility of end-to-end trainable two-stage u-net for detection of axillary lymph nodes in contrast-enhanced ct based on sparse annotations | |
| Zhang et al. | A new method for spherical object detection and its application to computer aided detection of pulmonary nodules in CT images | |
| Alolfe et al. | Development of a computer-aided classification system for cancer detection from digital mammograms | |
| Hasan et al. | MRI Brain Scans Classification Using Bi-directional Modified Gray Level Co-occurrence Matrix and Long Short-Term Memory | |
| Horsthemke et al. | Predicting LIDC diagnostic characteristics by combining spatial and diagnostic opinions |
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 |