KR102122073B1 - Parkinson's disease diagnosis apparatus using nigrosome-1 detected by machine learning, and method - Google Patents

Abstract

The apparatus for diagnosing Parkinson's disease based on the region of nigrosome 1 detected through machine learning according to the present invention, receives the first image related to multi echo size and phase from the MRI of the patient's brain. An image acquisition unit to acquire; An image processing unit post-processing the acquired first image to enable observation of the black matter and the nigrosome 1 region used as an image biomarker of Parkinson's disease; An image analysis unit analyzing the post-processed first image to classify a second image including the nigrosome 1 region, and detecting the nigrosome 1 region from the classified second image; And a diagnosis unit for analyzing the presence or absence of Parkinson's disease of the patient by analyzing whether the detected region of the nigrosome 1 is normal or not, wherein the image analysis unit analyzes the post-processed first image through machine learning. By doing so, a second image including the nigrosome 1 region may be classified.

Description

Apparatus and method for diagnosing Parkinson's disease based on the area of nigrosome 1 detected through machine learning{PARKINSON'S DISEASE DIAGNOSIS APPARATUS USING NIGROSOME-1 DETECTED BY MACHINE LEARNING, AND METHOD}

The present invention relates to an apparatus and method for diagnosing Parkinson's disease based on the region of nigrosome 1 detected through machine learning, and more specifically, analyzing magnetic resonance imaging (hereinafter referred to as'MRI') The present invention relates to a Parkinson's disease diagnosis apparatus and method for diagnosing Parkinson's disease.

Neurodegenerative disease refers to a disease that causes brain dysfunction as neurons disappear by some cause.

Typical neurodegenerative diseases include Alzheimer's disease, Parkinson's disease, and rarely Lou Gehrig's disease.

Among neurodegenerative diseases, Parkinson's disease is a representative neurodegenerative disease in which neurons are destroyed, and the body hardens, the hands and feet tremble, and the symptoms of inability to walk are accompanied by depression, anxiety, and significantly lower the quality of life.

As a method of diagnosing such a neurodegenerative disease, it is diagnosed by contact with mucous membranes, skin destruction, or a non-invasive method of diagnosis without internal body cavity other than natural or artificial body structures.

For example, Patent Document 1 and Patent Document 2 below disclose techniques for diagnosing neurodegenerative diseases according to the prior art.

Republic of Korea Patent Registration No. 10-1754291 (announced on July 6, 2017) Republic of Korea Patent Publication No. 10-2016-0058812 (released on May 25, 2016)

Sung-Min Gho et al., "Susceptibility Map-Weighted Imaging (SMWI) for Neuroimaging", Magnetic Resonance in Medicine 72:337-346 (2014) Yoonho Nam et al., "Imaging of Nigrosome 1 in Substantia Nigra at 3T Using Multiecho Susceptibility Map-Weighted Imaging (SMWI)", J. MAGN. RESON. IMAGING 2017; 46: 528-536 Christian Langkammer et al., "Quantitative susceptibility mapping (QSM) as a means to measure brain iron? A post mortem validation study", Neuroimage. 2012 Sep; 62(3):1593-9. doi: 10.1016/j.neuroimage. 2012.05.049. Epub 2012 May 24.

[18F]FP-CIT positron emission tomography (PET) using isotopes is the most objective method for diagnosing Parkinson's disease and differential diagnosis of drug-induced Parkinsonism.

However, the [18F]FP-CIT PET is a very expensive test method, and there is a risk of radiation exposure.

Therefore, there is a need to develop a technology capable of diagnosing Parkinson's disease by observing the region of nigrosome 1 using MRI.

An object of the present invention is to solve the problems as described above, to provide an apparatus and method for diagnosing Parkinson's disease by analyzing MRI to diagnose Parkinson's disease.

Another object of the present invention is to provide an apparatus and method for diagnosing Parkinson's disease by analyzing a region of nigrosome 1 presented as an image biomarker of Parkinson's disease.

In order to achieve the above object, the apparatus for diagnosing Parkinson's disease based on the region of nigrosome 1 detected through machine learning according to the present invention, from the MRI of the patient's brain, multi echo (multi echo ) An image acquisition unit that acquires a first image related to size and phase; An image processing unit post-processing the acquired first image to enable observation of the black matter and the nigrosome 1 region used as an image biomarker of Parkinson's disease; An image analysis unit analyzing the post-processed first image to classify a second image including the nigrosome 1 region, and detecting the nigrosome 1 region from the classified second image; And a diagnosis unit for analyzing the presence or absence of Parkinson's disease of the patient by analyzing whether the detected region of the nigrosome 1 is normal or not, wherein the image analysis unit analyzes the post-processed first image through machine learning. By doing so, a second image including the nigrosome 1 region may be classified.

In addition, the image analysis unit may designate a region for position detection on the classified second image, and detect the nigrosome 1 region within the designated region.

In addition, the image analysis unit, by using a region-convolutional neural network (Region-Convolutional Neural Network) method among the methods using the machine learning deep learning neural network (deep learnig neural network), the agent containing the region of the nigrosome 1 2 You can classify images.

In addition, the image analysis unit, by applying a convolutional neural network (CNN) to the post-processed first image, a feature map having a feature of a fully convolution layer (Fully convolution layer) Can be detected.

In addition, by applying a region proposal network (Region Proposal Network, RPN) to the feature map, a plurality of regions of interest (RoI) are set, and the set of regions of interest are resized to the same size. , Classified by applying the convolutional neural network (CNN) to the plurality of resized regions of interest using a support vector machine (SVM), classification loss and a bounding box for the classification result By processing the regression loss (bounding-box regression loss) according to a preset criterion, the second image including the nigrosome 1 region may be classified.

As described above, according to the apparatus and method for diagnosing Parkinson's disease based on the nigrosome 1 region detected through machine learning according to the present invention, MRI classifies only the image containing the nigrosome 1 region and classifies it. The effect of being able to diagnose the presence or absence of Parkinson's disease is obtained by analyzing the region of nigrosome 1 in the image.

And according to the present invention, by applying a susceptibility map weighted imaging protocol and a quantitative susceptibility mapping algorithm to improve the visibility of the nigrosome 1 region, and diagnose Parkinson's disease using images visualized with a black matter structure, universally supplied Parkinson's disease can be accurately diagnosed using MRI equipment, and the effect of improving the accuracy of the diagnosis results is obtained.

1 is a view illustrating an MRI including the black matter and the nigrosome 1 region,
Figure 2 is a block diagram of a Parkinson's disease diagnosis apparatus according to a preferred embodiment of the present invention,
3 is a view for explaining the operation of the image acquisition unit and the image processing unit,
4 and 5 are views for explaining the operation of the image analysis unit,
6 is a view for explaining the operation of the diagnostic unit,
7 is a view showing a state in which a diagnosis result is applied to an input image,
8 is a process step-by-step illustrating a method for diagnosing Parkinson's disease according to a preferred embodiment of the present invention.

Hereinafter, an apparatus and method for diagnosing Parkinson's disease according to a preferred embodiment of the present invention will be described in detail with reference to the accompanying drawings.

First, referring to FIG. 1, the criteria for diagnosing Parkinson's disease by observing the region of nigrosome 1 in the black matter will be described.

1 is a view illustrating an MRI including the black matter and the nigrosome 1 region.

FIG. 1(a) shows an imaging slab, and FIG. 1(b) shows an image of a normal person's black matter (substantia nigra) and the nigrosome 1 region, and FIG. 1(c) ) Shows images of the black matter and the nigrosome 1 region of Parkinson's disease patients.

The region of nigrosome 1 in the black matter of normal people is black as shown in FIG. 1(b), and the region of nigrosome 1 in patients with Parkinson's disease is relatively gray as shown in FIG. 1(c). Is displayed closely.

Therefore, the present invention can diagnose the presence or absence of Parkinson's disease by observing the shadow of the region of the Nigrosome 1 presented as an image biomarker of Parkinson's disease in MRI.

For example, the present invention applies the susceptibility map weighted imaging (hereinafter referred to as'SMWI') protocol in 3T MRI and a quantitative susceptibility mapping (hereinafter referred to as'QSM') algorithm therein. Parkinson's disease is diagnosed by analyzing the normality of the nigrosome 1 region by improving the visibility of the grosome 1 region.

The substantia nigra pars compacta is a midbrain structure that contains a dense population of dopamine neurons. These neurons progressively disappear in idiopathic Parkinson's disease (IPD), leading to disability. This region shows an increase in iron concentration in IPD patients compared to healthy controls.

Recently, as a result of visualizing a small portion of the dense part of the black matter known as nigrosome 1 in a high-resolution susceptibility contrast image of a healthy subject, the contrast between the nigrosome 1 and its surrounding substantia nigra regions was compared to the difference in iron concentration. As a result, the difference in susceptibility between the two sites was significantly decreased in patients with IPD.

The reduction in the susceptibility difference between these two sites is being used as an imaging biomarker for IPD.

Thus, the Nigrosome 1 structure has been successfully described in 7T MRI using high resolution (eg 0.3 mm planar resolution) T2-weighted imaging or susceptibility weighted imaging (SWI).

However, in a low magnetic field strength, such as 3T MRI, due to limited spatial resolution and signal/contrast-to-noise ratio (SNR/CNR), in 3D high-resolution T2-weighted imaging, a contrast, a structure in which contrast is significantly reduced, is observed.

This limitation hindered the reliability and applicability of Nigrosome 1 imaging in 3T MRI, despite several successful studies demonstrating the usefulness of the approach.

Recently, a new method for providing an improved magnetic susceptibility contrast has been proposed to solve the above problems.

One of them is to use multi-echo gradient recall echo (hereinafter referred to as'multi echo GRE') size image instead of using a single echo image to improve SNR. By combining, it shows relatively high accuracy in diagnosing IPD in 3T MRI.

An alternative to the size image is susceptibility-weighted imaging (SWI), which uses phase information as a weight mask to increase susceptibility contrast, or artifacts due to blooming of phase imaging. ).

Other susceptibility contrast related approaches are GRE phase (or frequency) images and quantitative susceptibility mapping (QSM), both of which have good susceptibility to susceptibility and have been widely applied in recent years.

In addition, a new method using a susceptibility weighting mask derived from QSM for a size image has been proposed.

This approach is similar to SWI, but it can solve SWI's blooming artifacts and potentially improve visualization of susceptibility changes.

QSM mask weighted imaging has proven useful in visualizing the structure of the Nigrosome 1.

The non-patent document 1 discloses an SMWI technology for neuroimaging, and the non-patent document 2 discloses an imaging technique of nigrosome 1 in black matter using multi-echo SMWI in 3T MRI, and the ratio Patent Document 3 discloses a QSM technology for brain iron measurement.

Therefore, the present invention diagnoses Parkinson's disease based on a change in the region of Nigrosome 1 due to Parkinson's disease according to the correlation between brain iron concentration and susceptibility.

2 is a block diagram of a Parkinson's disease diagnosis apparatus according to a preferred embodiment of the present invention.

Parkinson's disease diagnosis apparatus 10 according to a preferred embodiment of the present invention, as shown in Figure 2, an image acquisition unit 20, which acquires a multi-echo size and phase image from the MRI of the patient's brain, obtained The image processing unit 30 that post-processes the black matter and the nigrosome 1 region presented as an image biomarker of Parkinson's disease in the image, and analyzes the processed image to classify and classify the image containing the nigrosome 1 region. It includes an image analysis unit 40 for detecting a region of nigrosome 1 in an image and a diagnosis unit 50 for diagnosing Parkinson's disease by determining whether the region of nigrosome 1 is normal in each classified image.

The configuration of the image stroke feature and the image processor will be described in detail with reference to FIG. 3.

3 is a view for explaining the operation of the image acquisition unit and the image processing unit. FIG. 3 shows a process of generating an SMWI image for a structure of a nigrosome 1 from a multi-echo synthetic GRE image.

As shown in FIG. 3, the image acquisition unit 20 is communicatively connected to a MRI equipment 21 or a database (not shown) that stores and manages the MRI photographed by the MRI equipment 21 to prevent Parkinson's disease. Obtain the MRI of the patient to be diagnosed.

As shown in FIG. 3, the image processing unit 30 uses a multi-echo-size image and a multi-echo phase image composite image using a QSM algorithm for visualization of the nigrosome 1 structure. Creates SMWI video.

For example, the image processing unit 30 generates a channel-sized size image from the multi-channel composite image by the square root of the sum of squares of the multi-channel size image, and the phase image is a complex average after correcting the global phase offset of each channel. Combined (step 1).

Then, the image processing unit 30 combines six echo images into a single image by the square root of the sum of the image squares (step 2).

The image processor 30 calculates phase images of different TEs using a Laplacian unwrapping algorithm, and calculates a combined frequency w in each voxel (step 3).

The image processing unit 30 removes a background region from harmonic background phase removal using a Laplacian operator method in a frequency image (step 4).

Here, the QSM can be reconstructed using an improved sparse linear equation and least-squares (iLSQR) method.

For example, the error tolerance in the reconstruction parameters of iLSQR=0.01, the threshold D ₂ for an incomplete k-region mask, _{and thres} =0.1.

Subsequently, the image processing unit 30 additionally processes the resulting QSM to generate a QSM mask (S _mask ) for weighting compared to the susceptibility (step 5).

The mask may be generated using Equation 1.

[Equation 1]

Here, X is a quantitative susceptibility value (in ppm) calculated in step 4, and X _th is a paramagnetic threshold. The threshold can later be determined using Nigrosome 1 imaging data for optimal CNR.

Finally, the image processing unit 30 may generate an SMWI image by multiplying a multi-echo-combined sized image by a QSM mask using Equation 2 below.

[Equation 2]

Here, m is the number of multiplications with respect to the susceptibility weight, and mag is the multi-echo magnitude combined image of the two steps.

Next, the operation of the image analysis unit will be described in detail with reference to FIGS. 4 and 5.

4 and 5 are views for explaining the operation of the image analysis unit. FIG. 4 shows a process of classifying an image containing a region of nigrosome 1, and FIG. 5 shows a process of detecting a region of nigrosome 1 in the classified image.

As illustrated in FIG. 4 through the post-processing process of the image processing unit 30, a plurality of SWMI images may be acquired.

In general, MRI photographed for diagnosis of Parkinson's disease is about 40 to 70 sheets per patient, and among them, images containing a region of Nigrosome 1 used for diagnosing Parkinson's disease are within about 3 to 6 sheets.

As shown in FIG. 4, the image analysis unit 40 analyzes an image containing a region of nigrosome 1 among all MRIs through machine learning, and includes an image containing a region of nigrosome 1 and not included. You can classify images.

For example, the image analysis unit 40 is a Nairo using a region-convolutional neural network (hereinafter referred to as'RCNN') among the methods using a deep learning neural network of machine learning. A region including one region can be classified.

That is, the image analysis unit 40 detects a feature map using the obtained image using a convolutional neural network (hereinafter referred to as'CNN'), and a region proposal network (hereinafter referred to as a Region Proposal Network). About 2,000 regions of interest (RoI) are selected from the feature map detected using'RPN'), and after resizing the selected regions of interest to the same size, interest input to the CNN In order to increase the accuracy of the location of the region, a CNN is performed for each region of interest using a support vector machine (SVM) composed of neural networks, and classification loss and bounding boxes are classified for the classification results. By processing the regression loss (bounding-box regression loss), it can be minimized with only the image containing the region of the nigrosome 1 finally.

In addition, as shown in FIG. 5, the image analysis unit 40 designates a location where the region of the region of nigrosome 1 exists in each image including the region of nigrosome 1 classified through machine learning, and the designated nigrosome 1 Areas can be detected.

Next, the operation of the diagnostic unit will be described with reference to FIGS. 6 and 7.

6 is a diagram illustrating the operation of the diagnostic unit, and FIG. 7 is a diagram showing a state in which a diagnostic result is applied to an input image.

As illustrated in FIG. 6, the diagnostic unit 50 may diagnose whether Parkinson's disease is present or not by analyzing normality of the region of nigrosome 1 detected by the image analysis unit 40 through machine learning.

For example, as shown in (e) to (h) of FIG. 7, a diagnosis result may be applied according to the diagnosis result of Parkinson's disease in the input images of FIGS. 7A to 7D.

Next, a method of diagnosing Parkinson's disease according to a preferred embodiment of the present invention will be described in detail with reference to FIG. 8.

8 is a process diagram for explaining step-by-step a diagnosis method of Parkinson's disease according to a preferred embodiment of the present invention.

In step S10 of FIG. 8, the image acquisition unit 20 acquires a number of MRIs of patients who want to diagnose Parkinson's disease through communication with the MRI equipment 21 or a database.

Here, the image acquisition unit 20 may acquire a multi-echo GRE composite image in which a multi-echo size image and a multi-echo phase image are combined in 3T MRI.

In step S12, the image processing unit 30 post-processes the multi-echo GRE composite image in which the multi-echo size image and the multi-echo phase image are combined using the QSM algorithm to visualize the structure of the Nigrosome 1 SMWI. Create an image.

In step S14, the image analysis unit 40 analyzes an image containing a region of nigrosome 1 among all MRIs through machine learning to classify an image containing a region of nigrosome 1 and an image not included (classification). )do.

In addition, the image analysis unit 40 designates a location where the region of the region of nigrosome 1 exists in each image including the region of nigrosome 1 classified through machine learning, and detects the designated region of nigrosome 1 (S16).

Finally, the diagnosis unit 50 diagnoses the presence or absence of Parkinson's disease by judging whether or not the detected region of nigrosome 1 is normal in each image classified as an image including the nigrosome 1 region in the image analysis unit 40.

Through the above-described process, the present invention classifies only the image containing the region of nigrosome 1 in MRI, and analyzes the region of nigrosome 1 in the classified image to diagnose the presence or absence of Parkinson's disease.

Although the invention made by the present inventors has been described in detail according to the above-described embodiments, the present invention is not limited to the above-described embodiments, and it is needless to say that various modifications can be made without departing from the gist thereof.

The present invention is applied to an apparatus and method for diagnosing Parkinson's disease by diagnosing Parkinson's disease by classifying only the image containing the Nigrosome 1 region in MRI and analyzing the Nigrosome 1 region in the classified image.

10: Parkinson's disease diagnosis device
20: image acquisition unit
21: MRI equipment
30: image processing unit
40: image analysis unit
50: diagnostic unit

Claims (5)

An image acquisition unit acquiring a first image related to a multi echo size and phase from an MRI of a patient's brain;
An image processing unit post-processing the acquired first image to enable observation of the black matter and the nigrosome 1 region used as an image biomarker of Parkinson's disease;
An image analysis unit analyzing the post-processed first image to classify a second image including the nigrosome 1 region, and detecting the nigrosome 1 region from the classified second image; And
Including the diagnostic unit for analyzing the presence or absence of the detected Nigrosome 1 region to diagnose the presence or absence of Parkinson's disease in the patient;

The image analysis unit,
Parkinson's disease based on the area of nigrosome 1 detected through machine learning, characterized by classifying the second image containing the nigrosome 1 region by analyzing the post-processed first image through machine learning Device to diagnose.
According to claim 1,
The image analysis unit,
In the classified second image, a region for position detection is designated, and the nigrosome 1 region is detected within the designated region. Device for diagnosing illness.
According to claim 1,
The image analysis unit,
Among the methods using the deep learning neural network of the machine learning, using the Region-Convolutional Neural Network method, classifying the second image containing the region of the nigrosome 1 A device for diagnosing Parkinson's disease based on the region of nigrosome 1 detected through machine learning.
According to claim 3,
The image analysis unit,
A machine characterized by detecting a feature map having a feature of a fully convolution layer by applying a convolutional neural network (CNN) to the post-processed first image A device for diagnosing Parkinson's disease based on the region of nigrosome 1 detected through learning.
The method of claim 4,
A plurality of regions of interest (RoI) are set by applying a region proposal network (RPN) to the feature map,
Resizing the set plurality of regions of interest to the same size,
The support vector machine (SVM) is used to classify the convolutional neural network (CNN) by applying the convolutional neural network to the resized regions of interest.
The classification result and the bounding-box regression loss are processed according to a preset criterion with respect to the classification result, thereby classifying the second image including the nigrosome 1 region. A device for diagnosing Parkinson's disease based on the region of nigrosome 1 detected through machine learning.

Images