CN112562031B

CN112562031B - Nuclear magnetic resonance image clustering method based on sample distance reconstruction

Info

Publication number: CN112562031B
Application number: CN202011450094.4A
Authority: CN
Inventors: 潘晓光; 田奇; 宋晓晨; 马彩霞; 姚珊珊
Original assignee: Shanxi Sanyouhe Smart Information Technology Co Ltd
Current assignee: Shanxi Sanyouhe Smart Information Technology Co Ltd
Priority date: 2020-12-09
Filing date: 2020-12-09
Publication date: 2022-09-27
Anticipated expiration: 2040-12-09
Also published as: CN112562031A

Abstract

The invention belongs to the technical field of nuclear magnetic resonance image processing, and particularly relates to a nuclear magnetic resonance image clustering method based on sample distance reconstruction, which comprises the following steps: extracting a gray value of an original image; vectorizing the obtained gray matrix T; calculating Euclidean distance, Manhattan distance and Hamming distance of every two image vectors through the gray level vector to construct a distance matrix; merging the feature vectors into a feature matrix; reconstructing the distance between the samples; for the division of different MRI images. The invention can effectively solve the problem of insufficient information mining of the existing MRI images, and can quickly and accurately realize the clustering of the samples based on the distance between the reconstructed MRI image samples, thereby fully playing the role of the MRI technology in disease diagnosis. The method is used for clustering the nuclear magnetic resonance images.

Description

Nuclear magnetic resonance image clustering method based on sample distance reconstruction

Technical Field

The invention belongs to the technical field of nuclear magnetic resonance image processing, and particularly relates to a nuclear magnetic resonance image clustering method based on sample distance reconstruction.

Background

With the development of computer and medical diagnosis technology, the detection technology for human body focus has advanced greatly. Magnetic Resonance Imaging (MRI) has played a great role in the detection of tumors of the cardiovascular and cerebrovascular systems and the thoracic and abdominal cavities as a high-precision medical imaging technique. The main principle is that the nuclear magnetic resonance principle is utilized, the emitted electromagnetic waves are detected through an external gradient magnetic field according to different attenuations of the released energy in different structural environments in the material, so that the position and the type of the object atomic nucleus are formed, and the structural image in the object is drawn according to the position and the type.

At present, MRI is mainly involved in medical diagnosis in an auxiliary way or mainly depends on experience knowledge of doctors to judge the potential condition of a focus part in MRI imaging. In the study of MRI, based on the needs of clinical application and scientific research, the segmentation of MRI tumor images becomes one of the main ways to guide medical diagnosis, and researchers have proposed many segmentation methods of MRI tumor image brain and perform clustering of MRI images based on the segmentation methods, so as to identify the cancer stage of human organs and perform corresponding treatment. However, these methods require a priori knowledge of the type of cancer and the MRI image, and the results of segmenting the image have a large impact on the final classification and diagnosis.

Disclosure of Invention

Aiming at the technical problem that the existing MRI has larger influence on the final classification and diagnosis, the invention provides the nuclear magnetic resonance image clustering method based on sample distance reconstruction, which has strong stability, high accuracy and convenient use.

In order to solve the technical problems, the invention adopts the technical scheme that:

a nuclear magnetic resonance image clustering method based on sample distance reconstruction comprises the following steps:

s1: extracting a gray value of an original image based on an existing MRI tumor image dataset;

s2: vectorizing the obtained gray matrix T;

s3: calculating Euclidean distance, Manhattan distance and Hamming distance of every two image vectors through the gray level vector to construct a distance matrix;

s4: reconstructing the distance between the MRI images by fusing the three different distance matrixes obtained in the step S3, and merging the eigenvectors obtained by PCA transformation of the three distance matrixes into one eigenvector;

s5: converting the characteristic matrix into a new distance matrix DM by calculating Euclidean distances between rows in the characteristic matrix, and realizing the reconstruction of the distance between samples;

s6: and (4) obtaining a final distance result by executing hierarchical clustering, and realizing the division of different MRI images.

The method for extracting the gray value of the original image in S1 includes: let n existing MRI image data be m × m in size, and T represents the gray matrix of the original image.

The vectorization processing method in S2 includes: the matrix of m x m gray values is expanded from top to bottom and from left to right into m2 x 1 vectors, and the resulting column vectors are normalized by Z-score.

The method for constructing the distance matrix in the step S3 includes: respectively calculating Euclidean distance, Manhattan distance and Hamming distance of corresponding vectors of any two images, respectively representing all MRI to be evaluated in a distance matrix DM according to distance measurement by calculating the distance between the vectors corresponding to any two MRI images, respectively representing Euclidean distance, a distance matrix calculated by the Manhattan distance and the Hamming distance by using DM1, DM2 and DM3, wherein rows or columns of the distance matrix correspond to a sample, and elements in the distance matrix represent the distance between different MRI images.

The method for combining the feature vectors into one feature matrix in S4 includes: firstly, PCA conversion is carried out on each distance matrix, the first two features in the low-dimensional features for reserving the distance between the samples are selected and extracted, the extracted feature vectors are regarded as projection points of n MRI image samples in a two-dimensional plane space, so that the distance between the point pairs also reflects the distance between the samples, and then the feature vectors obtained by the PCA conversion of the three distance matrices obtained in the step S3 are combined into one feature matrix FM.

Compared with the prior art, the invention has the following beneficial effects:

the invention can effectively solve the problem of insufficient information mining of the existing MRI images, and can quickly and accurately realize the clustering of the samples based on the distance between the reconstructed MRI image samples, thereby fully playing the role of the MRI technology in disease diagnosis, helping doctors to quickly analyze the focus tumor information of patients and further establishing the subsequent treatment means. The prediction method is simple, convenient and effective, can ensure accurate clustering of a large number of MRI images, does not need image segmentation in advance so as to save time overhead, and can stably promote the distinction of the MRI images through the MRI sample set which is continuously amplified.

Drawings

FIG. 1 is a flow chart of the operation of the present invention;

FIG. 2 is a logic flow diagram of the algorithm of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

A nuclear magnetic resonance image clustering method based on sample distance reconstruction, as shown in fig. 1, includes the following steps:

step 1: extracting a gray value of an original image based on an existing MRI tumor image dataset;

step 2: vectorizing the obtained gray matrix T;

and 3, step 3: calculating Euclidean distance, Manhattan distance and Hamming distance of every two image vectors through the gray level vector to construct a distance matrix;

and 4, step 4: reconstructing the distance between the MRI images by fusing the three different distance matrixes obtained in the step 3, and merging the eigenvectors obtained by PCA transformation of the three distance matrixes into one eigenvector matrix;

and 5: and converting the characteristic matrix into a new distance matrix DM by calculating Euclidean distances between rows in the characteristic matrix, so as to realize the reconstruction of the distance between samples. As shown in fig. 2, based on the obtained feature matrix FM, the FM can be converted into a new distance matrix DM by calculating euclidean distances between rows in the FM, so as to reconstruct the distance between samples. Correspondingly, DM can be found to have the following properties: 1. the diagonal elements are all 0, and the off-diagonal elements are positive real numbers, representing the distance between different samples. 2. The size of the DM is n x n, and the rows or columns represent n MRI images.

Step 6: based on the obtained reconstructed distance matrix DM, as shown in fig. 2, a final distance result can be obtained by performing hierarchical clustering, which realizes the partition of different MRI images.

Further, the method for extracting the gray value of the original image in step 1 comprises the following steps: let n existing MRI image data be m × m, and T represents a gray matrix of the original image.

Further, the vectorization processing method in step 2 is as follows: the matrix of m x m gray values is expanded from top to bottom and from left to right into m2 x 1 vectors, and the resulting column vectors are normalized by Z-score.

Further, the method for constructing the distance matrix in step 3 comprises the following steps: as shown in fig. 2, based on the gradation vector V obtained for each MRI image, various distances between two image vectors V are calculated. Specifically, the principle is to calculate the euclidean distance, manhattan distance, and hamming distance of the corresponding vectors of any two images. By calculating the distance between the corresponding vectors of any two MRI images, all the MRI images to be evaluated can be represented in a distance matrix DM according to the distance measurement, and Euclidean distance, Manhattan distance and Hamming distance are respectively represented by DM1/DM2/DM3 to calculate the distance matrix. Wherein a row or a column of the distance matrix corresponds to one sample, and the elements in the distance matrix represent the distances between different MRI images. For example, DM1[ i, j ] represents the Euclidean distance between the ith and jth MRI images. As can be seen from the above, the calculated distance matrix has the following characteristics: 1. diagonal elements of the distance matrix are 0, while non-diagonal elements represent the distance between any two MRI images or the dissimilarity degree, and the value range is (0, 1); 2. the distance matrix has a size of n x n, and the rows or columns represent n MRI image samples.

Further, the method for combining the feature vectors into one feature matrix in step 4 is as follows: as shown in fig. 2, the distance between the MRI images is reconstructed by fusing the three different distance matrices based on the idea of ensemble learning with respect to the processed distance matrix. First, by performing PCA transformation on each distance matrix, low-dimensional features that preserve the distance between samples can be extracted, where the first two-dimensional features are selected for extraction. Since the extracted feature vectors can be regarded as projection points of the n MRI image samples in the two-dimensional plane space, the distance between the point pairs also reflects the distance between the samples. Therefore, the eigenvectors obtained by the PCA transformation of the three distance matrices are combined into one eigenvector matrix as the eigenvector matrix FM. Where FM is a matrix of size n x 6, and each row corresponds to one MRI image sample.

Although only the preferred embodiments of the present invention have been described in detail, the present invention is not limited to the above embodiments, and various changes can be made without departing from the spirit of the present invention within the knowledge of those skilled in the art, and all changes are encompassed in the scope of the present invention.

Claims

1. A nuclear magnetic resonance image clustering method based on sample distance reconstruction is characterized by comprising the following steps: comprises the following steps:

s2: vectorizing the obtained gray matrix T;

s3: calculating Euclidean distance, Manhattan distance and Hamming distance of every two image vectors through the gray level vectors, and constructing a distance matrix; the method for constructing the distance matrix in the step S3 includes: respectively calculating Euclidean distance, Manhattan distance and Hamming distance of vectors corresponding to any two images, respectively representing all the MRI to be evaluated in a distance matrix DM according to distance measurement by calculating the distance between the vectors corresponding to any two MRI images, respectively representing Euclidean distance by using DM1, DM2 and DM3, respectively representing distance matrixes calculated by using the Manhattan distance and the Hamming distance, wherein rows or columns of the distance matrixes correspond to a sample, and elements in the distance matrixes represent the distance between different MRI images;

s4: reconstructing the distance between the MRI images by fusing the three different distance matrixes obtained in the step S3, and merging the eigenvectors obtained by PCA transformation of the three distance matrixes into one eigenvector; the method for combining the feature vectors into one feature matrix in S4 includes: firstly, carrying out PCA conversion on each distance matrix, selecting and extracting the first two features in the low-dimensional features for reserving the distance between the samples, regarding the extracted feature vectors as projection points of n MRI image samples in a two-dimensional plane space, reflecting the distance between the samples by the distance between the point pairs, and then combining the feature vectors obtained by the PCA conversion of the three distance matrices obtained in the step S3 into a feature matrix FM;

s5: converting the characteristic matrix into a new distance matrix DM by calculating Euclidean distances between rows in the characteristic matrix to reconstruct the distance between samples;

s6: and obtaining a final distance result by performing hierarchical clustering, and realizing the division of different MRI images.

2. The method of claim 1, wherein the method comprises: the method for extracting the gray value of the original image in S1 includes: let n existing MRI image data be m × m in size, and T represents the gray matrix of the original image.

3. The method of claim 1, wherein the method comprises: the vectorization processing method in S2 includes: the matrix of gray values of m x m size is expanded from top to bottom and from left to right into vectors of m2 x 1 size, and the obtained column vectors are normalized by Z-score.