CN114428873B - Thoracic surgery examination data sorting method - Google Patents
Thoracic surgery examination data sorting method Download PDFInfo
- Publication number
- CN114428873B CN114428873B CN202210357195.XA CN202210357195A CN114428873B CN 114428873 B CN114428873 B CN 114428873B CN 202210357195 A CN202210357195 A CN 202210357195A CN 114428873 B CN114428873 B CN 114428873B
- Authority
- CN
- China
- Prior art keywords
- dictionary
- vector
- normal
- abnormal
- vectors
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/55—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/28—Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
- G16H30/20—ICT specially adapted for the handling or processing of medical images for handling medical images, e.g. DICOM, HL7 or PACS
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Library & Information Science (AREA)
- Databases & Information Systems (AREA)
- Health & Medical Sciences (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Epidemiology (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Primary Health Care (AREA)
- Public Health (AREA)
- Radiology & Medical Imaging (AREA)
- Probability & Statistics with Applications (AREA)
- Apparatus For Radiation Diagnosis (AREA)
- Measuring And Recording Apparatus For Diagnosis (AREA)
Abstract
The invention relates to the field of medical data, in particular to a thoracic surgery examination data sorting method, which comprises the following steps: performing K-SVD on a preset number of chest film images without diseases and diseases to obtain a normal dictionary, a normal sparse image set, an abnormal dictionary and an abnormal sparse image set, further obtaining the heat value of each dictionary vector in the normal dictionary and the abnormal dictionary, obtaining a normal dictionary vector set and an abnormal dictionary vector set according to the heat values of the normal dictionary, the abnormal dictionary and the dictionary vectors, performing K-SVD on all the chest film images, obtaining a final dictionary and a final sparse image set according to the normal dictionary vector set, the abnormal dictionary vector set and the heat value of each dictionary vector, and storing and classifying the final dictionary and the final sparse image set. The invention ensures that the chest radiography images are convenient to read, write and store after being sorted and the related data characteristics are quickly inquired, so that doctors can quickly analyze the state of an illness, and the examination receiving efficiency of hospitals is improved.
Description
Technical Field
The invention relates to the field of medical data, in particular to a thoracic surgery examination data sorting method.
Background
When a hospital performs physical examination on a patient or a doctor detects chest diseases and respiratory diseases of the patient, a chest radiograph of the patient is usually acquired, a large number of chest radiographs exist in a hospital database along with long-time receiving of a hospital, the chest radiographs are thoracic surgery examination data, and the chest radiographs have important significance for the follow-up examination of the patient and the big data analysis and mining of medical data. Generally, a large number of chest radiography images need to be queried, read and stored to obtain the desired chest radiography data characteristics, but due to the large quantity of chest radiography data and the lack of arrangement and analysis of the data, the query, reading and storage efficiency is low, and particularly when certain data characteristics in a large number of chest radiography images need to be obtained and analyzed, such as the situation of analyzing pulmonary effusion at different periods, the obtaining efficiency of the data characteristics is low, so that doctors want to take a needle in the open sea when obtaining the desired data characteristics.
Based on the above, a large amount of chest radiography data needs to be sorted, and the data sorting refers to operations such as checking, coding, storing, classifying and the like on the data, so that a large amount of chest radiography data is convenient to read, write, store and quickly inquire related data characteristics after sorting, a doctor can quickly analyze the state of an illness, for example, quickly visualize chest radiography of a patient, analyze change of a focus of the patient and the like, and the efficiency of receiving a doctor in a hospital is increased.
Disclosure of Invention
In order to solve the above technical problems, the present invention aims to provide a method for collating thoracic surgery examination data, which adopts the following technical scheme:
the invention provides a thoracic surgery examination data sorting method, which comprises the following steps:
performing K-SVD on a preset number of chest film images without diseases to obtain a normal dictionary and a normal sparse image set, and performing K-SVD on the preset number of chest film images without diseases to obtain an abnormal dictionary and an abnormal sparse image set;
obtaining the heat value of each dictionary vector in the normal dictionary according to the normal sparse image set and the normal dictionary, and obtaining the heat value of each dictionary vector in the abnormal dictionary according to the abnormal sparse image set and the abnormal dictionary;
obtaining a normal dictionary vector set and an abnormal dictionary vector set according to the heat values of the normal dictionary, the abnormal dictionary and the dictionary vectors, performing K-SVD on all the chest image without disease and all the chest image with disease, and obtaining a final dictionary and a final sparse image set according to the normal dictionary vector set, the abnormal dictionary vector set and the heat value of each dictionary vector in the set;
And storing the final dictionary, classifying the images in the final sparse image set, then storing, and directly inquiring the classified final sparse images when the chest image data needs to be inquired.
Further, the step of obtaining the heat value of each dictionary vector in the normal dictionary and the heat value of each dictionary vector in the abnormal dictionary comprises:
firstly, normalizing each normal sparse image in a normal sparse image set, then summing all normal sparse images after normalization to obtain a normal fusion image, wherein each pixel in the normal fusion image corresponds to one dictionary vector in a normal dictionary, and finally normalizing the normal fusion image, wherein the gray value of each pixel on the normal fusion image after normalization is regarded as the heat value of the dictionary vector corresponding to each pixel;
in a similar way, each abnormal sparse image in the abnormal sparse image set is subjected to normalization processing, then all abnormal sparse images after normalization processing are summed to obtain an abnormal fusion image, each pixel in the abnormal fusion image corresponds to one dictionary vector in the abnormal dictionary, finally the abnormal fusion image is subjected to normalization processing, and the gray value of each pixel on the abnormal fusion image after normalization processing is regarded as the heat value of the dictionary vector corresponding to each pixel.
Further, the method for acquiring the normal dictionary vector set and the abnormal dictionary vector set comprises the following steps:
for all dictionary vectors in the normal dictionary, acquiring a dictionary vector set with the heat value larger than a first preset threshold value, and calling the dictionary vector set as a first normal category, and acquiring a dictionary vector set with the heat value not larger than the first preset threshold value, and calling the dictionary vector set as a second normal category;
for all dictionary vectors in the abnormal dictionary, acquiring a dictionary vector set with the heat value larger than a first preset threshold value, which is called a first abnormal category, and acquiring a dictionary vector set with the heat value not larger than the first preset threshold value, which is called a second abnormal category;
and then all dictionary vectors which belong to the second abnormal category but do not belong to the second cross category are obtained, and are called as an abnormal dictionary vector set, and the first cross category is called as a normal dictionary vector set.
Further, the method for obtaining the final dictionary comprises the following steps:
firstly, normalizing the heat values of all dictionary vectors in a normal dictionary vector set, normalizing the heat values of all dictionary vectors in an abnormal dictionary vector set, combining all dictionary vectors in the normal dictionary vector set and all dictionary vectors in the abnormal dictionary vector set into a dictionary vector set, and calling the dictionary vectors in a reference dictionary as reference vectors;
In the SVD process of all the chest film images without diseases and all the chest film images with diseases, a random dictionary is initialized randomly, dictionary vectors in the random dictionary are called as vectors to be updated, and the vectors to be updated in the random dictionary correspond to reference vectors in a reference dictionary one by one;
and then updating each vector to be updated in the random dictionary by using a K-SVD algorithm, generating a reference displacement for each vector to be updated after each vector to be updated is updated once, updating each vector to be updated according to the K-SVD algorithm after each vector to be updated generates a reference displacement, continuously repeating the updating process until all vectors to be updated are converged, regarding the random dictionary after the vectors to be updated are converged as a final dictionary, and simultaneously obtaining a final sparse image set of all the chest radiography images according to the K-SVD algorithm.
Further, the method for acquiring the reference displacement comprises the following steps:
obtaining a reference vector corresponding to a vector to be updated, calculating a difference vector between the reference vector and the vector to be updated, obtaining a direction pointed by the difference vector, calculating a product of a modular length of the difference vector and a heat value corresponding to the reference vector, and reconstructing a vector according to the direction by taking the product as the modular length, wherein the vector is a reference displacement corresponding to the vector to be updated.
Further, the step of obtaining the first cross category and the second cross category comprises:
selecting two dictionary vectors from a first normal category and a first abnormal category respectively as a dictionary vector pair, and calculating Euclidean distance of the dictionary vector pair, wherein if the Euclidean distance is smaller than a second preset threshold value, the dictionary vector pair can form a similar vector, and if the Euclidean distance is not smaller than the second preset threshold value, the dictionary vector pair cannot form a similar vector;
then all dictionary vector pairs which can form similar vectors in the first normal category and the first abnormal category are obtained, and a dictionary vector set formed by all the obtained dictionary vector pairs is called as a first cross category;
and similarly, a second cross category is obtained by using the same method according to the second normal category and the second abnormal category.
Further, the step of acquiring the normal dictionary and the normal sparse image set includes:
flattening each chest film image without disease into a one-dimensional vector, inputting the one-dimensional vectors corresponding to the preset number of chest film images without disease into a K-SVD algorithm, wherein the K-SVD algorithm outputs a normal dictionary which is a collection of dictionary vectors;
The K-SVD algorithm also outputs a preset number of sparse vectors which correspond to the preset number of chest radiography images without diseases one by one, each sparse vector is mapped into an image which is called a sparse image, and the set of all sparse images is called a normal sparse image set;
and in the same way, obtaining an abnormal dictionary and an abnormal sparse image set according to the preset number of the chest film images with diseases.
The embodiment of the invention has the following beneficial effects: according to the invention, the normal dictionary and the abnormal dictionary are obtained by performing K-SVD on the chest radiography images without diseases and the chest radiography images with diseases, the heat value of the dictionary vector is calculated, the reference dictionary capable of clearly representing the disease characteristics of the patients with diseases is obtained according to the heat values of the normal dictionary, the abnormal dictionary and the dictionary vector, and all the chest radiography image data are compressed and data are sorted according to the reference dictionary, so that a large amount of chest radiography data are convenient to read, write and store after being sorted, and the related data characteristics are quickly inquired, so that doctors can quickly analyze the disease conditions, and the hospital receiving efficiency is increased.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions and advantages of the prior art, the drawings used in the embodiments or the description of the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
Fig. 1 is a flowchart of a thoracic surgery examination data sorting method according to an embodiment of the present invention.
Detailed Description
To further illustrate the technical means and functional effects of the present invention adopted to achieve the predetermined objects, the following detailed description will be given for the thoracic surgery examination data organizing method according to the present invention, and the specific implementation, structure, features and functional effects thereof with reference to the accompanying drawings and the preferred embodiments. In the following description, the different references to "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
The following describes a specific scheme of the thoracic surgery examination data sorting method provided by the present invention in detail with reference to the accompanying drawings.
Referring to fig. 1, a flow chart of a thoracic surgery examination data sorting method according to an embodiment of the present invention is shown, the method includes the following steps:
And S001, performing K-SVD on the affected and unaffected chest radiography images.
For a large number of chest film images, each chest film image is a single-channel gray-scale image, the images need to be checked firstly, and the specific method is that a doctor marks whether the chest film has a diseased condition when shooting the chest film, and rejects unclear images, so that the chest film with the diseased condition is classified into one type, and the chest film without the diseased condition is classified into another type; in the invention, when the breast image compression result can obviously show the sick condition, the compression result has stronger characteristic representation capability, if the breast image compression result can not clearly show the sick condition, the compression result has weaker characteristic representation capability, and the conventional K-SVD algorithm can possibly lead to weak compression result characteristic representation capability and can not respectively show that the compression result has the sick condition, even the diseased characteristic data in the compression results may be lost.
In order to solve the problem of poor characteristic representation capability of a compression result, firstly, randomly acquiring N chest radiography image data without diseases from a large number of chest radiography images, wherein the value of N is one tenth of the total number of all the chest radiography images, and then carrying out K-SVD decomposition on the N chest radiography images without diseases to obtain a normal dictionary and a normal sparse image set, wherein the specific method comprises the following steps:
flattening each chest film image without disease into a one-dimensional vector, and recording the flattened one-dimensional vector of the ith chest film image without disease as
If M is the pixel number of the chest picture image, N one-dimensional vectors are correspondingly obtained, the N one-dimensional vectors are input into a K-SVD algorithm, the K-SVD algorithm outputs a dictionary, the dictionary is called a normal dictionary, the normal dictionary is a set formed by Q dictionary vectors, Q is set to be 512 multiplied by 512, each dictionary vector is M-dimensional, and the Q dictionary vector in the normal dictionary is recorded as
In addition, the K-SVD algorithm also outputs N sparse vectors which are in one-to-one correspondence with the N one-dimensional vectors, and the ith sparse vector is recorded as
According to the K-SVD algorithm, it can be known that,
i.e. vectors
Is the linear superposition of dictionary vectors in a normal dictionary, and the superposition coefficient of the qth dictionary vector is 
(ii) a According to the K-SVD algorithm, dictionary vectors in a normal dictionary represent all the characteristics of the N chest picture images, wherein the dictionary vectors in the normal dictionary represent all the characteristics of the N chest picture images
A feature on the chest radiograph image, such as a wheel well feature that may represent the thorax, and a disease feature that includes features in the absence of disease and features in the presence of different diseases, but for which it is unknown what features are specifically represented; however, it is to be appreciated that: if it is used
The larger the size, the more attention is paid to the ith chest picture image
Characteristic features, or aspects, expressed
The more important the represented features.
The invention also represents sparse vectors in the form of images, e.g. the ith sparse vector
Can be expressed as a size of
Is called a normal sparse image
Thus, therefore, it is
Correspond to
The above-mentioned normal sparse image set refers to a set of N normal sparse images.
In view of the above
Is a dictionary vector
The coefficient of superposition of (a) and (b),
the larger, the description
The more important the characteristic represented, the
Correspond to
Above, the normal sparse image can be considered as a normal sparse image
The higher the gray value of the pixel is, the more important and more interesting the corresponding dictionary vector is.
Similarly, N chest radiography image data with diseases are randomly acquired from a large number of chest radiography images, and then the N chest radiography images with diseases are subjected to K-SVD to obtain an abnormal dictionary and an abnormal sparse image set.
Because the normal dictionary is obtained by calculation from the chest radiography image without the disease, it can be determined that the dictionary vectors in the normal dictionary all describe some features when the disease is absent, and the larger the gray value of the pixel on the normal sparse image in the normal sparse image set is, the more important the dictionary vector corresponding to the pixel is. Because the abnormal dictionary is obtained by calculation from the chest film image with the disease, it can be determined that the dictionary vector in the abnormal dictionary can describe some features when the disease is present in addition to some features when the disease is absent, and the larger the gray value of the pixel on the abnormal sparse image in the abnormal sparse image set is, the more important the dictionary vector corresponding to the pixel is.
And S002, obtaining a heat value of each dictionary vector in the normal dictionary and the abnormal dictionary according to the normal sparse image set, the normal dictionary, the abnormal sparse image set and the abnormal dictionary.
Firstly, each normal sparse image in the normal sparse image set is normalized, for example, the normalization result of the q pixel on the ith normal sparse image is 
In which
Representing the gray value of the q-th pixel on the ith normal sparse image,
representing the sum of all pixel gray values on the ith normal sparse image.
Summing all the normalized normal sparse images to obtain a normal fusion image, wherein each pixel value on all the normal sparse images corresponds to one dictionary vector in a normal dictionary, so that each pixel in the fusion image corresponds to one dictionary vector in the normal dictionary, and finally normalizing the normal fusion image, wherein the gray value of each pixel on the normalized normal fusion image is regarded as the heat value of the dictionary vector corresponding to each pixel; the larger the heat value of the dictionary vector is, the more important the features represented by the dictionary vector are on all the images of the chest radiograph without diseases, and the features represented by the dictionary vector are necessarily the features without diseases, such as the features of the thorax contour, the features of the ribs and the like.
Similarly, each abnormal sparse image in the abnormal sparse image set is normalized, then all the abnormal sparse images after normalization are summed to obtain an abnormal fusion image, each pixel in the abnormal fusion image corresponds to one dictionary vector in the abnormal dictionary, finally the abnormal fusion image is normalized, the gray value of each pixel on the abnormal fusion image after normalization is regarded as the heat value of the dictionary vector corresponding to each pixel, the greater the heat value of the dictionary vector obtained here, the more important the feature represented by the dictionary vector is on all the affected chest radiography images, and the feature represented by the dictionary vector may be the feature without disease and may also be the feature in disease, such as the shadow feature caused by lung effusion.
And S003, acquiring a normal dictionary vector set and an abnormal dictionary vector set according to all dictionary vectors in the normal dictionary and the abnormal dictionary and the heat values of all the dictionary vectors.
Regarding dictionary vectors in the normal dictionary, the dictionary vectors with the heat value larger than the threshold th1 are called as a first normal category, and the dictionary vectors with the heat value not larger than the threshold th1 are called as a second normal category; the first normal category represents some of the most important features without disease, the second normal category represents some of the less important features without disease, such as some features introduced by image noise or some features due to patient-to-patient variation, etc., the th1= 0.5.
Regarding dictionary vectors in the abnormal dictionary, the dictionary vectors with the heat value larger than the threshold th1 are called as a first abnormal category, and the dictionary vectors with the heat value not larger than the threshold th1 are called as a first abnormal category; the first abnormal category represents some of the most important features when diseased, and the second normal category represents some features that are not important when diseased, such as some features introduced by image noise or due to patient-to-patient variation, some features introduced due to the presence of a lesion, and so on.
Obtaining a first cross category according to the first normal category and the first abnormal category, wherein the cross category obtaining method comprises the following steps: selecting two dictionary vectors from the first normal category and the first abnormal category respectively as a dictionary vector pair, and calculating Euclidean distance of the dictionary vector pair, wherein if the Euclidean distance is less than th2, the dictionary vector pair can form a similar vector, and if the Euclidean distance is not less than th2, the dictionary vector pair can not form a similar vector.
And then all dictionary vector pairs which can form similar vectors in the first normal category and the first abnormal category are obtained, and a dictionary vector set formed by all the obtained dictionary vector pairs is called as a first cross category.
The invention provides a th2 acquisition method, which comprises the following steps: in the first normal category, euclidean distances between a certain dictionary vector and all other dictionary vectors are acquired, the minimum value of the euclidean distances is taken as the nearest neighbor distance of the dictionary vector, the nearest neighbor distances of all dictionary vectors are calculated, and one half of the nearest neighbor distances is taken as th 2.
If the dictionary vectors in the first normal category and the first abnormal category are regarded as a distribution area of a high-dimensional space, and the dictionary vectors in the first cross category belong to an intersection area of a space where the first normal category is located and a space area where the first abnormal category is located, the first cross category can be regarded as an intersection of the first normal category and the first abnormal category; therefore, the dictionary vector in the first cross category can represent the image characteristics of the chest film without illness and the characteristics of the chest film with illness, so that the invention considers that the dictionary vector in the first cross category is the lung characteristics of the normal disease-free area of the lung of the patient except the characteristics of the illness area; since the medical science focuses more on which areas of the lung of a patient with diseases are diseased and which areas are non-diseased, the method for obtaining the first cross category has an important role in subsequent data arrangement, can increase the feature characterization capability of the compression result of a subsequent chest image, and the first cross category is called as a normal dictionary vector set.
And similarly, a second cross category is obtained according to a second normal category and a second abnormal category, the second cross category represents some features introduced by image noise or some features caused by individual differences of patients, and all dictionary vectors which belong to the second abnormal category and do not belong to the second cross category are obtained, are called an abnormal dictionary vector set and represent some features of the lung lesion area of the patient with the disease, and can increase the feature characterization capability of the compression result of the subsequent chest radiography image.
And S004, obtaining a final dictionary and a final sparse image set according to the normal dictionary vector set and the abnormal dictionary vector set.
Normalizing the heat values of all dictionary vectors in the normal dictionary vector set, normalizing the heat values of all dictionary vectors in the abnormal dictionary vector set, combining the dictionary vectors in the normal dictionary vector set and the dictionary vectors in the abnormal dictionary vector set into a dictionary vector set, regarding the dictionary vector set as a reference dictionary, and regarding the dictionary vectors in the reference dictionary as reference vectors, wherein the reference dictionary represents the characteristics of the lung lesion area and the characteristics of the lung normal area of the sick patient, and the characteristics are important bases for judging and analyzing the disease condition by doctors.
All the chest film images without diseases and all the chest film images with diseases are combined into a chest film image set, and then K-SVD decomposition is carried out on the images in the chest film image set by utilizing a K-SVD algorithm.
In the K-SVD decomposition process, a random dictionary is initialized randomly at first, a dictionary vector in the random dictionary is called as a vector to be updated, and the vector to be updated in the random dictionary corresponds to a reference vector in a reference dictionary one by one; then, each vector to be updated in the random dictionary is continuously updated by utilizing a K-SVD algorithm, specifically, the kth vector to be updated in the random dictionary is used
For the purpose of illustration, when the K-SVD algorithm pair
After one time of updating, the invention makes
Then, a reference displacement is generated
Then the vector is to be updated
Become to stand for moreNew vector
Then the K-SVD algorithm pair treats the updated vector again
Updating is carried out, and the updating process is continuously repeated until the vector to be updated
Converging; similarly, repeating the updating process for all other vectors to be updated in the random dictionary at the same time until all vectors to be updated in the random dictionary converge, wherein the converged random dictionary is called as a final dictionary; and simultaneously, obtaining a final sparse image set of the chest radiography image set according to a K-SVD algorithm, wherein the final sparse image set is a compression result of the chest radiography image set.
Reference displacement
The acquisition method comprises the following steps: obtaining a vector to be updated
Corresponding reference vector
Calculating a reference vector
And the vector to be updated
Difference vector of (2)
Obtaining a vector
Angle of direction pointed to
Calculating a vector
Die length of
And a reference vector
Corresponding heat value
Product of (2)
To do so by
Is a die length of
For direction, a vector is reconstructed, and the vector is the reference displacement corresponding to the vector to be updated
The reference displacement
Can guide the vector to be updated
Continuously approaching to the reference vector
Gradually approaching the features characterized in the random dictionary to the features characterized in the reference dictionary, wherein
Reference vector is explained
The more important, the more will be the need to approach the reference vector
。
Compared with the conventional K-SVD algorithm, the method provided by the invention adds the updating step of reference displacement, and can ensure that the obtained final dictionary contains the characteristics of the lung lesion area and the characteristics of the normal area of the lung of the patient, so that all chest image compression results have stronger characteristic characterization capability, the normal characteristics and the diseased characteristics of the lung can be clearly embodied, and the compression results are prevented from being doped with too many other characteristics unrelated to disease analysis.
And S005, arranging the chest image data according to the compression result of the chest image.
And storing the final dictionary, classifying all sparse images in the final sparse image set, and storing the classified sparse images, wherein the specific classification method comprises the following steps: PCA dimension reduction is carried out on all sparse images, all sparse images are reduced to P dimension, P =256 is carried out in the invention, then all dimension reduction results are classified by utilizing a K-means algorithm, the invention is divided into B types, the value of B in the invention is determined according to the types of lung diseases, B =24 in the invention, sparse images in the same type have similar disease condition characteristics, the sparse images belonging to the same type are stored together, the classified query of different disease conditions is facilitated, and meanwhile, due to the fact that the sparse images occupy small storage space, the calculation amount is small when the sparse images participate in the operation, the chest radiography images are classified, stored and sorted according to the method, the reading, the storage and the rapid query of related disease condition data characteristics are facilitated, doctors can carry out rapid analysis on the disease conditions, and the hospital receiving efficiency is increased.
It should be noted that: the sequence of the above embodiments of the present invention is only for description, and does not represent the advantages or disadvantages of the embodiments. And that specific embodiments have been described above. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.
All the embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from other embodiments.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and should not be taken as limiting the scope of the present invention, which is intended to cover any modifications, equivalents, improvements, etc. within the spirit and scope of the present invention.
Claims (5)
1. A method for collating thoracic surgical examination data, comprising the steps of:
performing K-SVD on a preset number of chest radiography images without diseases to obtain a normal dictionary and a normal sparse image set, and performing K-SVD on the preset number of chest radiography images without diseases to obtain an abnormal dictionary and an abnormal sparse image set;
acquiring a heat value of each dictionary vector in the normal dictionary according to the normal sparse image set and the normal dictionary, and acquiring a heat value of each dictionary vector in the abnormal dictionary according to the abnormal sparse image set and the abnormal dictionary;
obtaining a normal dictionary vector set and an abnormal dictionary vector set according to the normal dictionary, the abnormal dictionary and the heat value of the dictionary vector, performing K-SVD (K-singular value decomposition) on all the chest film images without diseases and all the chest film images with diseases, and obtaining a final dictionary and a final sparse image set according to the normal dictionary vector set, the abnormal dictionary vector set and the heat value of each dictionary vector in the sets;
The method for acquiring the normal dictionary vector set and the abnormal dictionary vector set comprises the following steps:
for all dictionary vectors in the normal dictionary, acquiring a dictionary vector set with a heat value larger than a first preset threshold value, called a first normal category, and acquiring a dictionary vector set with a heat value not larger than the first preset threshold value, called a second normal category; for all dictionary vectors in the abnormal dictionary, acquiring a dictionary vector set with a heat value larger than a first preset threshold value, called a first abnormal category, and acquiring a dictionary vector set with a heat value not larger than the first preset threshold value, called a second abnormal category; obtaining a first cross category according to the first normal category and the first abnormal category, obtaining a second cross category according to the second normal category and the second abnormal category, then obtaining all dictionary vectors which belong to the second abnormal category but do not belong to the second cross category, and calling all the dictionary vectors as an abnormal dictionary vector set, and calling the first cross category as a normal dictionary vector set;
the method for acquiring the final dictionary comprises the following steps:
firstly, normalizing the heat values of all dictionary vectors in a normal dictionary vector set, normalizing the heat values of all dictionary vectors in an abnormal dictionary vector set, combining all dictionary vectors in the normal dictionary vector set and all dictionary vectors in the abnormal dictionary vector set into a dictionary vector set, and calling the dictionary vectors in a reference dictionary as reference vectors;
In the SVD process of all the chest film images without diseases and all the chest film images with diseases, a random dictionary is initialized randomly, dictionary vectors in the random dictionary are called as vectors to be updated, and the vectors to be updated in the random dictionary correspond to reference vectors in a reference dictionary one by one;
updating each vector to be updated in the random dictionary by using a K-SVD algorithm, enabling each vector to be updated to generate a reference displacement after each vector to be updated is updated once, updating each vector to be updated according to the K-SVD algorithm after each vector to be updated generates a reference displacement, continuously repeating the updating process until all vectors to be updated are converged, regarding the random dictionary after the vectors to be updated are converged as a final dictionary, and simultaneously obtaining a final sparse image set of all chest radiography images according to the K-SVD algorithm;
and storing the final dictionary, classifying the images in the final sparse image set, then storing, and directly inquiring the classified final sparse images when the chest image data needs to be inquired.
2. The thoracic surgery examination data collating method according to claim 1, wherein the obtaining step of the heat value of each dictionary vector in the normal dictionary and the heat value of each dictionary vector in the abnormal dictionary includes:
Firstly, normalizing each normal sparse image in a normal sparse image set, then summing all normal sparse images after normalization to obtain a normal fusion image, wherein each pixel in the normal fusion image corresponds to one dictionary vector in a normal dictionary, and finally normalizing the normal fusion image, wherein the gray value of each pixel on the normal fusion image after normalization is regarded as the heat value of the dictionary vector corresponding to each pixel;
in a similar way, each abnormal sparse image in the abnormal sparse image set is subjected to normalization processing, then all abnormal sparse images after normalization processing are summed to obtain an abnormal fusion image, each pixel in the abnormal fusion image corresponds to one dictionary vector in the abnormal dictionary, finally the abnormal fusion image is subjected to normalization processing, and the gray value of each pixel on the abnormal fusion image after normalization processing is regarded as the heat value of the dictionary vector corresponding to each pixel.
3. The thoracic surgical examination data interpretation method of claim 1, wherein the reference displacement is obtained by:
obtaining a reference vector corresponding to a vector to be updated, calculating a difference vector between the reference vector and the vector to be updated, obtaining a direction pointed by the difference vector, calculating a product of a modular length of the difference vector and a heat value corresponding to the reference vector, and reconstructing a vector according to the direction by taking the product as the modular length, wherein the vector is a reference displacement corresponding to the vector to be updated.
4. The thoracic surgical examination data interpretation method of claim 1, wherein the first cross category and the second cross category are obtained by:
selecting two dictionary vectors from a first normal category and a first abnormal category respectively as a dictionary vector pair, and calculating Euclidean distance of the dictionary vector pair, wherein if the Euclidean distance is smaller than a second preset threshold value, the dictionary vector pair can form a similar vector, and if the Euclidean distance is not smaller than the second preset threshold value, the dictionary vector pair cannot form a similar vector;
then, all dictionary vector pairs which can form similar vectors in the first normal category and the first abnormal category are obtained, and a dictionary vector set formed by all the obtained dictionary vector pairs is called as a first cross category;
and similarly, a second cross category is obtained by using the same method according to the second normal category and the second abnormal category.
5. The thoracic surgical examination data interpretation method of claim 1, wherein the step of obtaining the normal dictionary and the normal sparse image set comprises:
flattening each chest film image without disease into a one-dimensional vector, inputting the one-dimensional vectors corresponding to the preset number of chest film images without disease into a K-SVD algorithm, wherein the K-SVD algorithm outputs a normal dictionary which is a collection of dictionary vectors;
The K-SVD algorithm also outputs a preset number of sparse vectors, the preset number of sparse vectors correspond to the preset number of chest images one by one, each sparse vector is mapped into an image, the image is called a sparse image, and the set of all sparse images is called a normal sparse image set;
and in the same way, obtaining an abnormal dictionary and an abnormal sparse image set according to the preset number of the chest film images with diseases.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210357195.XA CN114428873B (en) | 2022-04-07 | 2022-04-07 | Thoracic surgery examination data sorting method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210357195.XA CN114428873B (en) | 2022-04-07 | 2022-04-07 | Thoracic surgery examination data sorting method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114428873A CN114428873A (en) | 2022-05-03 |
| CN114428873B true CN114428873B (en) | 2022-06-28 |
Family
ID=81314282
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210357195.XA Active CN114428873B (en) | 2022-04-07 | 2022-04-07 | Thoracic surgery examination data sorting method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114428873B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103914823A (en) * | 2012-12-31 | 2014-07-09 | 复旦大学 | Method for registering stereoscopic medical image rapidly, accurately and non-linearly based on sparse representation |
| CN107730479A (en) * | 2017-08-30 | 2018-02-23 | 中山大学 | High dynamic range images based on compressed sensing go artifact fusion method |
| CN111667453A (en) * | 2020-04-21 | 2020-09-15 | 浙江工业大学 | Gastrointestinal endoscope image anomaly detection method based on local feature and class mark embedded constraint dictionary learning |
| CN113112557A (en) * | 2021-02-23 | 2021-07-13 | 西南科技大学 | Block self-adaptive carton image compression method based on dictionary learning |
| CN113223032A (en) * | 2021-04-27 | 2021-08-06 | 武汉纺织大学 | Double-sparse decomposition-based complex image Canny edge detection method |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104200451B (en) * | 2014-08-28 | 2017-05-10 | 西北工业大学 | Image fusion method based on non-local sparse K-SVD algorithm |
| CN105574475B (en) * | 2014-11-05 | 2019-10-22 | 华东师范大学 | A kind of rarefaction representation classification method based on common vector dictionary |
| US10610162B2 (en) * | 2016-05-31 | 2020-04-07 | Stmicroelectronics S.R.L. | Method for the detecting electrocardiogram anomalies and corresponding system |
| CN106845551B (en) * | 2017-01-24 | 2020-08-11 | 湘潭大学 | Tissue pathology image identification method |
| CN107194912B (en) * | 2017-04-20 | 2020-12-29 | 中北大学 | Brain CT/MR image fusion method based on sparse representation and improved coupled dictionary learning |
| US10866298B2 (en) * | 2017-08-03 | 2020-12-15 | Siemens Healthcare Gmbh | Low rank and spatial regularization model for magnetic resonance fingerprinting |
| US20190095787A1 (en) * | 2017-09-27 | 2019-03-28 | Hsiang Tsung Kung | Sparse coding based classification |
-
2022
- 2022-04-07 CN CN202210357195.XA patent/CN114428873B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103914823A (en) * | 2012-12-31 | 2014-07-09 | 复旦大学 | Method for registering stereoscopic medical image rapidly, accurately and non-linearly based on sparse representation |
| CN107730479A (en) * | 2017-08-30 | 2018-02-23 | 中山大学 | High dynamic range images based on compressed sensing go artifact fusion method |
| CN111667453A (en) * | 2020-04-21 | 2020-09-15 | 浙江工业大学 | Gastrointestinal endoscope image anomaly detection method based on local feature and class mark embedded constraint dictionary learning |
| CN113112557A (en) * | 2021-02-23 | 2021-07-13 | 西南科技大学 | Block self-adaptive carton image compression method based on dictionary learning |
| CN113223032A (en) * | 2021-04-27 | 2021-08-06 | 武汉纺织大学 | Double-sparse decomposition-based complex image Canny edge detection method |
Non-Patent Citations (2)
| Title |
|---|
| 《Brain tumor classification using sparse coding and dictionary learning》;Saif Dawood Salman Al-Shaikhli;《IEEE》;20150129;第1-5页 * |
| 基于结构联合字典的肺部LDCT图像降噪;代晓婷等;《电子学报》;20180615(第06期);第1445-1453页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114428873A (en) | 2022-05-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Da Nobrega et al. | Lung nodule malignancy classification in chest computed tomography images using transfer learning and convolutional neural networks | |
| Houssein et al. | Deep and machine learning techniques for medical imaging-based breast cancer: A comprehensive review | |
| Yousef et al. | A holistic overview of deep learning approach in medical imaging | |
| Helwan et al. | Deep networks in identifying CT brain hemorrhage | |
| Kim et al. | Deep feature learning for pulmonary nodule classification in a lung CT | |
| Šajn et al. | Image processing and machine learning for fully automated probabilistic evaluation of medical images | |
| Chanu et al. | Retracted article: computer-aided detection of brain tumor from magnetic resonance images using deep learning network | |
| Khatibi et al. | Proposing feature engineering method based on deep learning and K-NNs for ECG beat classification and arrhythmia detection | |
| Ma et al. | A new classifier fusion method based on historical and on-line classification reliability for recognizing common CT imaging signs of lung diseases | |
| Depeursinge et al. | Comparative performance analysis of state-of-the-art classification algorithms applied to lung tissue categorization | |
| CN114863225A (en) | Image processing model training method, image processing model generation device, image processing equipment and image processing medium | |
| CN113989576A (en) | Medical image classification method combining wavelet transformation and tensor network | |
| CN114428873B (en) | Thoracic surgery examination data sorting method | |
| Manojlović et al. | Deep embedded clustering algorithm for clustering PACS repositories | |
| CN114224354B (en) | Arrhythmia classification method, arrhythmia classification device, and readable storage medium | |
| Hasan et al. | Science and business | |
| Valério et al. | Deepmammo: deep transfer learning for lesion classification of mammographic images | |
| CN112560784B (en) | Electrocardiogram classification method based on dynamic multi-scale convolutional neural network | |
| Ghose et al. | A breast cancer detection model using a tuned svm classifier | |
| Amin et al. | Semi-supervised learning for limited medical data using generative adversarial network and transfer learning | |
| Tan et al. | Graph-regularized manifold-aware conditional Wasserstein GAN for brain functional connectivity generation | |
| Prusty et al. | Scalar invariant transform based deep learning framework for detecting heart failures using ECG signals | |
| Akbar et al. | PENet: Continuous-Valued Pulmonary Edema Severity Prediction On Chest X-ray Using Siamese Convolutional Networks | |
| CN112989971A (en) | Electrocardiogram data fusion method and device for different data sources | |
| Tawfeeq et al. | Predication of Most Significant Features in Medical Image by Utilized CNN and Heatmap. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |