CN111899206A - Medical brain image fusion method based on convolutional dictionary learning - Google Patents

Abstract

The invention discloses a medical brain image fusion method based on convolutional dictionary learning, which comprises the following steps: step 1, decomposing a source medical brain image into a low-frequency component and a high-frequency component, and step 2, fusing the low-frequency component; and 3, fusing the high-frequency components. Step 4, obtaining fused high-frequency components through fast Fourier inverse transformation; and 5, carrying out image reconstruction on the fused low-frequency component and the fused high-frequency component to obtain a fused image. The invention has the advantages that: better image information retention, advanced effect on visual quality and objective index, clearer boundaries of brain tissues, frontal sinuses and the like in the obtained image, and more contribution to the diagnosis and analysis of illness state of medical staff.

Description

Medical brain image fusion method based on convolutional dictionary learning

Technical Field

The invention relates to the technical field of image fusion, in particular to a medical brain image fusion method based on convolutional dictionary learning.

Background

Multi-modal medical brain images play a very important role in medical diagnosis. With the rapid development of medical imaging technology, it becomes possible to obtain human anatomy and functional description with high resolution and larger information amount. This development has prompted research in the field of medical image analysis. At present, various medical brain images have respective characteristics, and the fusion of the brain images is helpful for accurate diagnosis and can also be used for electronic medical treatment. Therefore, medical brain image fusion techniques for obtaining high information quality and compact information representation have attracted attention in the field of information synthesis and enhancement.

Imaging methods such as Computed Tomography (CT) have high spatial resolution and geometric features, and can clearly show bone structures. Magnetic Resonance Imaging (MRI) reveals soft tissues and organs. Therefore, the combination of CT and MRI images provides more information on the pathological state of the relevant organ, improving the diagnostic ability for clinical applications. In recent years, scholars at home and abroad propose various fusion methods and strategies: 1) transform domain based methods such as DWT, NSCT, etc.; 2) fusion methods based on sparse domains, such as sparse representation, joint sparse representation, adaptive sparse representation, and the like, and variants thereof; 3) the fusion strategy based on the neural network, such as the convolution neural network, the generative confrontation network and other classical network models. The various fusion algorithms obtain better fusion effect under certain specific conditions, but ignore semantic conflict of medical brain images, namely, the brightness of the CT image represents the density of the tissues, and the brightness of the MR-T2 image represents the mobility and magnetism of the tissues.

Disclosure of Invention

Aiming at the defects in the prior art, the invention provides a medical brain image fusion method based on convolutional dictionary learning, and the defects in the prior art are overcome.

In order to realize the purpose, the technical scheme adopted by the invention is as follows:

a medical brain image fusion method based on convolutional dictionary learning comprises the following steps:

step 1, source medical brain image { s }_A,s_BDecomposed into low-frequency components by fast Fourier transform

And high frequency component

And 2, fusing the low-frequency components by using an average rule:

wherein,

is a low frequency fused component.

Step 3, high frequency componentThe fusion uses maximum rule fusion:

wherein,

and

respectively representing high frequency components

High-frequency sparse coefficients obtained by Convolutional Basis Pursuit Denoising (CBPDN),

is the sparse coefficient of the high-frequency fused component, | · | > count₁Represents the 1-norm of.

In the step 4, the step of,

obtaining fused high-frequency component H through fast Fourier inverse transformation_SF；

Step 5, mixing

And

the image reconstruction is carried out as follows:

obtaining a fused image S_F。

Compared with the prior art, the invention has the advantages that:

better image information retention, advanced effect on visual quality and objective index, clearer boundaries of brain tissues, frontal sinuses and the like in the obtained image, and more contribution to the diagnosis and analysis of illness state of medical staff.

Drawings

FIG. 1 is a block diagram of a medical brain image fusion method according to an embodiment of the present invention;

fig. 2 is a source image of a medical brain tested in an embodiment of the present invention, which is divided into four different groups of brain images (a), (b), (c), and (d), the upper row is CT images, and the lower row is MRI images;

FIG. 3 is a comparison graph of the fusion results of a group of images according to example (a) of the present invention;

FIG. 4 is a comparison graph of the fusion results of the group of images according to example (b) of the present invention;

FIG. 5 is a comparison graph of the fusion results of the group of images according to example (c) of the present invention;

FIG. 6 is a comparison graph of the fusion results of the group of images in example (d) of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail below with reference to the accompanying drawings by way of examples.

As shown in fig. 1, the medical brain image fusion method based on the convolutional dictionary learning includes the following steps:

step 1, source medical brain image { s }_A,s_BDecomposed into low-frequency components by fast Fourier transform

And high frequency component

Fusing the components respectively by using the following rules;

and 2, fusing the low-frequency components by using an average rule:

wherein,

is a low frequency fused component.

Step 3, fusing the high-frequency components by using a maximum rule:

wherein,

and

respectively representing high frequency components

High-frequency sparse coefficients obtained by Convolutional Basis Pursuit Denoising (CBPDN),

is the sparse coefficient of the high-frequency fused component, | · | > count₁Represents the 1-norm of.

In the step 4, the step of,

obtaining fused high-frequency component H through fast Fourier inverse transformation_SF；

Step 5, mixing

And

carrying out image reconstruction formula (1) to obtain a fused image S_F。

Results and analysis of the experiments

To verify the superiority of the proposed algorithm, all algorithms were applied to 4 sets of classical medical brain images (as shown in fig. 2). CT and MRI are taken as examples: CT images employ an X-ray beam that passes directly through the brain, with the beam being somewhat blunted as it exits the other side, since the beam strikes dense tissue during passage, and the blunting or "attenuation" of the X-rays comes from the tissue density encountered along the way. Very dense tissue (e.g., bone) can block a large amount of X-rays. Gray matter will block somewhat, while fluid will block even less; MRI images place protons (here brain protons) in a magnetic field, they are able to receive and then transmit electromagnetic energy, the intensity of the transmitted energy being proportional to the number of protons in the tissue, the signal intensity varying due to the properties of the microenvironment of each proton (e.g., its mobility and local homogeneity of the magnetic field), and the MRI signals may be "weighted" to emphasize certain properties but not others.

(1) Fused image result graph

The fusion results of 4 groups of medical brain images on various methods are shown in fig. 3-6. (a) Representing fusion results based on an Adaptive Sparse Representation (ASR); (b) representing fusion results based on discrete wavelet transform and sparse representation (DWT-SR-4); (c) indicating the fusion results presented herein. And combining 4 groups of fusion results, the fused image obtained by DWT-SR-4 has the worst quality, and the fused image not only has a blocky artifact structure (see FIG. 4(b)), but also has serious damage to the fused image information (see FIG. 5 (b)). The ASR fusion result is superior to the former, but compared with the method provided by the inventor, the overall brightness is lower, and the method is not beneficial to the diagnosis of doctors at the later stage.

(2) Objective evaluation

To better verify the performance of the proposed algorithm, 3Q-series objective evaluation indexes (Q)^e，Q^TEAnd Q^p) Acts on the comparative process and the process of the invention (as shown in tables 1-4). By combining the analysis of tables 1-4, the fusion algorithm designed by the invention is slightly lower than an ASR algorithm in part of objective evaluation indexes, but is obviously higher than DWT-SR-4. Q of the proposed algorithm^e，Q^TEAnd Q^pHas an average value of 0.4112, 0.4712 and 0.4982, and compared with a DWT-SR-4 fusion algorithm, the method disclosed by the invention has the value of Q^e，Q^TEAnd Q^pThe upper parts are respectively improved by 9.62 percent, 7.42 percent and 43.73 percent.

Table 1(a) Objective evaluation results of group images on all fusion methods

Table 2(b) Objective evaluation results of group images on all fusion methods

Table 3 Objective evaluation results of group (c) images on all fusion methods

Table 4 Objective evaluation results of group (d) images on all fusion methods

It will be appreciated by those of ordinary skill in the art that the examples described herein are intended to assist the reader in understanding the manner in which the invention is practiced, and it is to be understood that the scope of the invention is not limited to such specifically recited statements and examples. Those skilled in the art can make various other specific changes and combinations based on the teachings of the present invention without departing from the spirit of the invention, and these changes and combinations are within the scope of the invention.

Claims (1)

1. A medical brain image fusion method based on convolutional dictionary learning is characterized by comprising the following steps:

step 1, source medical brain image { s }_A,s_BDecomposed into low-frequency components by fast Fourier transform

And high frequency component

And 2, fusing the low-frequency components by using an average rule:

wherein,

is a low frequency fused component;

and 3, fusing the high-frequency components by using a maximum rule:

wherein,

and

respectively representing high frequency components

High-frequency sparse coefficients obtained through convolution basis pursuit denoising,

is the sparse coefficient of the high-frequency fused component, | · | > count₁A 1-norm representing ·;

in the step 4, the step of,

obtaining fused high-frequency component H through fast Fourier inverse transformation_SF；

Step 5, mixing

And

the image reconstruction is carried out as follows:

obtaining a fused image S_F。

Images