KR102164091B1 - PET image analysis method for diagnosis of dementia - Google Patents

Abstract

,

,

으로 각각 정의하고,
상기 시상면 영상에 대하여 정상, 초기, 치매 출력값을

,

,

로 각각 정의하고,
상기 관상면 영상에 대하여 정상, 초기, 치매 출력값을

,

,

로 각각 정의하는 경우,
상기 축상면 영상, 상기 시상면 영상, 및 상기 관상면 영상 각각의 출력값에 대하여 평균값을 아래의 식 (1), (2), (3) 같이 계산하고,

(1)

(2)

(3)
아래의 식 (4)와 같은 기준으로 진단 결과를 산출하며,

(4)
여기서,

는 최종 진단결과로써 0 이면 정상, 1 이면 초기, 2 이면 치매로 분류된다. max()는 원소 중에서 최대값인 것을 특징으로 한다.
본 발명에 따른 치매 진단 및 예측을 위한 심층학습 기반 PET 영상 분석 방법을 실시하는 경우 단순히 뇌의 특정 영상만에 기초한 종래의 진단 방법과 대비하여 보다 입체적이고 정밀한 치매 진단이 가능하다는 이점이 있다.The present invention relates to a PET (positron emission tomography) image analysis method for diagnosing dementia, and more specifically, a sagittal plane and a coronal view from an axial plane image obtained by a PET image obtained by tomography of the axial plane of the brain. After automatically generating a coronal plane image, it is analyzed by automatically classifying and analyzing normal, early, intermediate, late or normal, dementia early, and dementia stages for dementia by using artificial intelligence learning results on these three cross-sectional images. It relates to a PET image analysis method based on deep learning for the diagnosis and prediction of possible dementia.
The present invention generates a plurality of axial plane images by photographing an axial plane of the brain in a tomography, and then image interpolation is performed on the plurality of axial plane images to generate a 3D model, and the 3D model The sagittal and coronal planes of are sampled at regular intervals to generate a plurality of sagittal images and a plurality of coronal images, and learning is performed using a machine learning method. As a PET image analysis method having a reasoning device for performing a diagnosis of dementia that performs a process of inferring a degree,
Providing a predetermined axial plane image consisting of a plurality of;
Generating a plurality of predetermined sagittal images and coronal images, respectively, based on the predetermined axial plane image using a 3D model generator;
The predetermined axial plane image, the predetermined sagittal plane image, and the predetermined coronal plane image are provided to an axial plane reasoning machine, a sagittal plane reasoning machine, a coronal plane reasoning machine, and the axial plane reasoning machine, a sagittal plane reasoning machine, and Comprising the step of performing a diagnosis of dementia by inputting the result of the coronal surface inferior to a selector,

Normal, initial, dementia output values for the axial plane image

,

,

Respectively defined as,
Normal, initial, and dementia output values for the sagittal image

,

,

Respectively defined as,
Normal, initial and dementia output values for the coronal image

,

,

If each is defined as,
The average value of the output values of the axial plane image, the sagittal plane image, and the coronal plane image is calculated as shown in the following equations (1), (2), and (3),

(One)

(2)

(3)
The diagnosis result is calculated based on the criteria shown in Equation (4) below,

(4)
here,

Is the final diagnosis result, and if it is 0, it is classified as normal, if it is 1, it is classified as dementia. max() is characterized by being the maximum value among the elements.
When the PET image analysis method based on deep learning for diagnosing and predicting dementia according to the present invention is performed, there is an advantage in that it is possible to diagnose dementia more three-dimensionally and accurately compared to a conventional diagnosis method based only on a specific image of the brain.

Description

PET image analysis method for diagnosis of dementia based on deep learning for diagnosis and prediction of dementia

The present invention relates to a PET (positron emission tomography) image analysis method for diagnosing dementia, and more specifically, a sagittal plane and a coronal view from an axial plane image obtained by a PET image obtained by tomography of the axial plane of the brain. After automatically generating a coronal plane image, it is analyzed by automatically classifying and analyzing normal, early, intermediate, late or normal, dementia early, and dementia stages for dementia by using artificial intelligence learning results on these three cross-sectional images. It relates to a PET image analysis method based on deep learning for the diagnosis and prediction of possible dementia.

Dementia is the biggest challenge in modern life science and medicine.

According to US statistics, the mortality rate from major diseases such as cancer and cardiovascular disease is steadily decreasing due to improved treatment, but it is reported that the only disease that is increasing mortality is dementia.

In fact, according to the American Alzheimer's Association, the death rate from dementia from 2000 to 2014 soared 89%. This is because the incidence of dementia is increasing along with the aging population, and it is difficult to come up with a countermeasure as no clear cause has been identified.

According to the “National Research and Development Survey and Analysis Report for Dementia” released in 2015 by the Korea Health Industry Development Institute, the number of dementia patients is expected to triple from 46.8 per 1 million in 2015 to 131.5 in 2050.

As for the cause of dementia, only a few clues are known as shadows.

A typical example is the fact that a lump (plaque) of a protein called amyloid beta around the brain cells of dementia patients is observed as a kind of waste product.

It is believed that this waste product causes necrosis of nerve cells. As this phenomenon is known, research on treatments to remove amyloid beta, the causative agent of plaque, has been in rapid success in the last 20 years.

The problem is that despite these studies, no one has yet found a dementia solution. Research director Choi said, "There are many cases where cognitive abilities do not improve even with drugs that remove amyloid beta plaques, and on the contrary, there are cases in which dementia does not occur even though a lot of plaque is accumulated."

Naturally, drug development is also slow. There are four drugs approved by the U.S. Food and Drug Administration (FDA) (5 including cocktail drugs), but it is not a fundamental treatment only as a symptom reliever that helps overall brain activity by supplementing brain signaling substances.

Due to the slow performance, drug development was abandoned one after another.

Seoul National University School of Medicine Professor Mok In-hee also said, "The Alzheimer's disease drug development ranking has been pushed from 6th in 2015 to 8th in the debate hosted by the National Academy of Science and Technology of Korea" on December 8, 2017. I mentioned it.

In recent years, instead of digging into only amyloid beta protein, the atmosphere is looking for the cause of dementia from various angles.

Shin Hee-seop, director of the Research Center for Cognitive and Social Studies of the Institute of Basic Sciences, said, "As the amyloid beta research has not been successful, many researchers are now turning their attention to another waste product, "tau protein."

The tau protein is an object of interest after the amyloid beta protein.

Originally, it is a material constituting microfibers in nerve cells, but when it is transformed by amyloid beta, it becomes a waste product and aggregates. If amyloid beta builds up outside the neuron and causes problems, the tau protein builds up inside the neuron. As such, it is difficult to penetrate with drugs, and more difficult to remove than amyloid.

Meanwhile, dementia is also known to be deeply related to genes.

Mice that have removed the TREM2 gene, which is believed to be related to dementia, have many immune cells that are active when there is a problem in the brain (white is immune cells, picture on the left), but mice with the gene are relatively inactive.

There are also studies that have completely changed their perspectives.

An attempt to understand in terms of inflammation combined with immunology is typical.

As waste proteins such as amyloid travel through blood from other organs such as the kidneys to the brain, attacking normal proteins like a zombie and infecting them, studies are underway to investigate this process.

Immune cells that act as police officers to protect the brain are also known to be denatured during this process.

In addition, genes related to dementia are in the process of revealing genome translation studies.

Currently, about three genes, such as TREM2, are attracting attention.

Meanwhile, the association with intestinal bacteria is also being discussed.

Although the direct relationship between dementia and intestinal bacteria has not yet been revealed, it is drawing attention as the association with other brain diseases such as autism is revealed.

Therefore, it can be said that the prevention and treatment of dementia to improve the quality of life of the elderly is one of the very challenging tasks.

According to a report on ‘Early Diagnosis of Dementia,’ published by the Centers for Disease Control and Prevention, the prevalence of dementia is rapidly increasing due to the rapid aging of the population, and medical expenses are increasing significantly.

The prevalence of dementia among the elderly over 65 was 10.18% (72 million) in 2017.

After that, it is expected to increase to 10.39% (84 million) in 2020 and 15.06% (2.17 million) in 2050.

In addition, dementia is a disease that incurs significant long-term care and medical expenses, so the cost of managing dementia is also increasing.

According to the National Assembly Budget Office, the cost of managing dementia in 2013 was estimated at 11 trillion won, reaching 1% of GDP (Gross Domestic Product), and by 2050 it will quadruple to 43.2 trillion won, reaching 1.5% of GDP. do.

Among these, there is currently no cure to cure dementia, but early detection is becoming more important as medication (symptom reliever) can delay the onset of dementia.

It is reported that the hospitalization rate in nursing facilities decreases by 55% after 5 years of drug treatment from the initial stage of dementia.

An official from the Centers for Disease Control and Prevention said, "Because dementia gradually declines in cognitive function 10 to 20 years before the onset, a new early diagnosis method is required."

“A test method for delaying and preventing the occurrence of dementia in patients with mild cognitive impairment should be developed.” In addition, he added, “With the advent of new technology using human resources related to dementia, clinical information, and brain imaging information <see PET Figure 1> It is also necessary to develop possible new drugs.”

In addition, there is an analysis that the economic and social costs of about 3 trillion won can be saved by drug treatment in the early diagnosis of dementia.

Therefore, early diagnosis of dementia can be viewed as very important in reality.

One of the methods widely used in the industry for early diagnosis of dementia is using PET imaging.

However, in the conventional image analysis using PET images, an expert (a doctor or a medical image analyst) selects and analyzes about 5 images from a plurality of PET images, and diagnoses dementia manually.

In other words, for the diagnosis of dementia, an average of 110 PET images are taken with a tomography of an axial plane. Among 110 tomographic PET images, one image of the middle layer (eg, the 55th layer) and Two images of each of the upper and lower layers (eg, the 45th, 50th, 60th, and 65th layers) close to the middle layer are being used to analyze dementia.

For this reason, the results of pre-diagnosis of dementia using PET images of the related art have a lot of lack of reliability.

1. Patent application number; 1020160066266, Title of invention: Early dementia diagnosis method using 13C magnetic resonance spectroscopy

1. F. Li, L. Tran, K.-H. Thung, S. Ji, D. Shen, and J. Li, “A robust deep model for improved classification of AD/MCI patients,” IEEE J. Biomed. Health Inform., vol. 19, no. 5, pp. 1610-1616, 2015. 2. A. Payan and G. Montana, “Predicting alzheimer's disease: a neuroimaging study with 3D convolutional neural networks,” arXiv:1502.02506 [cs.CV], 2015. 3. E. Bron, M. Smits, W. van der Flier et al., "Standardized evaluation of algorithms for computer-aided diagnosis of dementia based on structural MRI: The CADDementia challenge," NeuroImage, 2015. 4. S. Plis, D. Hjelm, R. Salakhutdinov, E. Allen, H. Bockholt, J. Long, H. Johnson, J. Paulsen, J. Turner, and V. Calhoun, “Deep learning for neuroimaging: a validation study,” Frontiers in Neuroscience, vol. 8, 2014.

1. The present invention is proposed to solve such a conventional problem, and not only uses all PET images, but also automatically generates sagittal and coronal images based on the axial plane, and then based on all of these images, the state of the brain We propose a method that can automatically diagnose the progress of dementia by analyzing in three dimensions.

2. Securing the accuracy of dementia diagnosis through deep learning analysis of 3D PET images

For the diagnosis of dementia, about 100 slices of images can be obtained per user, but since it is difficult for a doctor (human) to check all of the PET images of about 100 slices and diagnose dementia, currently, only 5 slices are used to diagnose dementia. However, by performing a 3D PET image analysis applying deep learning, that is, a correlation analysis between image segmentation and slices, we intend to diagnose dementia more accurately.

3. In the encoder-decoder network such as U-Net, which is mainly used for segmentation and analysis of medical images, parameters such as the size of the convolution window and the number of layers are selected as trial and error. We propose a method of diagnosing dementia by systematically selecting deep learning parameters by applying an optimization algorithm such as et al.

Embodiment 1

The present invention generates a plurality of axial plane images by photographing an axial plane of the brain in a tomography, and then image interpolation is performed on the plurality of axial plane images to generate a 3D model, and the 3D model The sagittal and coronal planes of are sampled at regular intervals to generate a plurality of sagittal images and a plurality of coronal images, and learning is performed using a machine learning method. As a PET image analysis method having a reasoning device for performing a diagnosis of dementia that performs a process of inferring a degree,

Providing a predetermined axial plane image consisting of a plurality of;

Generating a plurality of predetermined sagittal images and coronal images, respectively, based on the predetermined axial plane image using a 3D model generator;

The predetermined axial plane image, the predetermined sagittal plane image, and the predetermined coronal plane image are provided to an axial plane reasoning machine, a sagittal plane reasoning machine, a coronal plane reasoning machine, and the axial plane reasoning machine, a sagittal plane reasoning machine, and Comprising the step of performing a diagnosis of dementia by inputting the result of the coronal surface inferior to a selector,

,

,

으로 각각 정의하고, For the axial plane image, the normal, initial, and dementia output values output from the axial plane inference machine

,

,

Respectively defined as,

,

,

로 각각 정의하고, The normal, initial, and dementia output values output from the sagittal plane inferred for the sagittal plane image

,

,

Respectively defined as,

,

,

로 각각 정의하는 경우, The normal, initial, and dementia output values output from the coronal surface inferior for the coronal surface image

,

,

If each is defined as,

The selector calculates an average value for each output value of the axial plane image, the sagittal plane image, and the coronal plane image as shown in Equations (1), (2), and (3) below,

(1)

(One)

(2)

(2)

(3)

(3)

The diagnosis result is calculated based on the criteria shown in Equation (4) below,

(4)

(4)

는 최종 진단결과로써 0 이면 정상, 1 이면 초기, 2 이면 치매로 분류된다. max()는 원소 중에서 최대값인 것을 특징으로 한다.here,

Is the final diagnosis result, and if it is 0, it is classified as normal, if it is 1, it is classified as dementia. max() is characterized by being the maximum value among the elements.

Example 2

The present invention generates a plurality of axial plane images by photographing an axial plane of the brain in a tomography, and then image interpolation is performed on the plurality of axial plane images to generate a 3D model, and the 3D model The sagittal and coronal planes of are sampled at regular intervals to generate a plurality of sagittal images and a plurality of coronal images, and learning is performed using a machine learning method. As a PET image analysis method having a reasoning device for performing a diagnosis of dementia that performs a process of inferring a degree,

Providing a predetermined axial plane image consisting of a plurality of;

Generating a plurality of predetermined sagittal images and coronal images, respectively, based on the predetermined axial plane image using a 3D model generator;

The predetermined axial plane image, the predetermined sagittal plane image, and the predetermined coronal plane image are provided to an axial plane reasoning machine, a sagittal plane reasoning machine, a coronal plane reasoning machine, and the axial plane reasoning machine, a sagittal plane reasoning machine, and Comprising the step of performing a diagnosis of dementia by inputting the result of the coronal surface inferior to a selector,

,

,

으로 각각 정의하고, For the axial plane image, the normal, initial, and dementia output values output from the axial plane inference machine

,

,

Respectively defined as,

,

,

로 각각 정의하고, The normal, initial, and dementia output values output from the sagittal plane inferred for the sagittal plane image

,

,

Respectively defined as,

,

,

로 각각 정의하는 경우, The normal, initial, and dementia output values output from the coronal surface inferior for the coronal surface image

,

,

If each is defined as,

The selector calculates a calculated average value for the axial plane image, the sagittal plane image, and the coronal plane image as in Equation (5)-Equation (7),

(5)

(5)

(6)

(6)

(7)

(7)

Output the diagnosis result as in Equation (8) below,

(8)

(8)

는 최종 진단결과로써 0 이면 정상, 1 이면 초기, 2 이면 치매로 분류하는 것을 특징으로 한다.here,

Is characterized in that 0 is normal, 1 is early, and 2 is dementia as the final diagnosis result.

Example 3

The present invention generates a plurality of axial plane images by photographing an axial plane of the brain in a tomography, and then image interpolation is performed on the plurality of axial plane images to generate a 3D model, and the 3D model The sagittal and coronal planes of are sampled at regular intervals to generate a plurality of sagittal images and a plurality of coronal images, and learning is performed using a machine learning method. As a PET image analysis method having a reasoning device for performing a diagnosis of dementia that performs a process of inferring a degree,

Providing a predetermined axial plane image consisting of a plurality of;

Generating a plurality of predetermined sagittal images and coronal images, respectively, based on the predetermined axial plane image using a 3D model generator;

The predetermined axial plane image, the predetermined sagittal plane image, and the predetermined coronal plane image are provided to an axial plane reasoning machine, a sagittal plane reasoning machine, a coronal plane reasoning machine, and the axial plane reasoning machine, a sagittal plane reasoning machine, and Comprising the step of performing a diagnosis of dementia by inputting the result of the coronal surface inferior to a selector,

,

,

으로 각각 정의하고, For the axial plane image, the normal, initial, and dementia output values output from the axial plane inference machine

,

,

Respectively defined as,

,

,

로 각각 정의하고, The normal, initial, and dementia output values output from the sagittal plane inferred for the sagittal plane image

,

,

Respectively defined as,

,

,

로 각각 정의하는 경우, The normal, initial, and dementia output values output from the coronal surface inferior for the coronal surface image

,

,

If each is defined as,

The selector is characterized in that the diagnosis of dementia is based on a maximum value among output values of the axial, sagittal and coronal images.

When the PET image analysis method based on deep learning for diagnosing and predicting dementia according to the present invention is performed, there is an advantage in that it is possible to diagnose dementia more three-dimensionally and accurately compared to a conventional diagnosis method based only on a specific image of the brain.

1 is a view for explaining a process of acquiring a sagittal plane and a coronal plane image from an axial plane image.
2 is a diagram for explaining a process of learning and inferring on an image for each section.
3 is a diagram for explaining a process of diagnosing the degree of dementia according to the present invention when an axial image of a specific examinee is provided.

Hereinafter, a PET image analysis method for diagnosis of dementia proposed by the present invention will be described with reference to the drawings.

The present invention automatically generates sagittal plane and coronal plane images from PET images taken of the axial plane of the brain in a tomography, and uses artificial intelligence learning results in three cross-sectional images for dementia. Proposed a method of automatically classifying normal, early, intermediate, late or normal, early dementia, and dementia stages.

One. Sagittal and Coronal surface Image acquisition process

First, a process of generating a sagittal image and a coronal image, which is a longitudinal cross-sectional image of the brain, from an axial plane PET image obtained by taking a tomography of a horizontal cross section of the brain is shown in FIG. 1.

The axial PET image is photographed at regular slice intervals, and the photographed image is stored in DICOM (digital imaging and communication in medicine), a standard for digital medical image transmission, which is a medical image standard.

The axial plane PET image conforming to the DICOM standard contains the information of the slice spacing.

In the present invention, image interpolation is performed on the axial PET image to generate a 3D model, and the sagittal and coronal planes of the 3D model are sampled at regular intervals Acquire a surface image.

As for the image interpolation method, bilinear interpolation or cubic spline may be applied.

2. Video learning method for automatic classification

A process of learning about the photographed PET axial plane and the sagittal plane and coronal plane images generated based on the images, and inferring using the learning or training results are shown in FIG. 2.

Each of the axial, sagittal, and coronal images is trained using a machine learning method, and inference is performed using the results of the learning.

The machine learning method can apply Adaboost (adaptive boosting), SVM (support vector machine), random forest, and deep learning techniques.

Meanwhile, a weight according to the training result will be set and fixed in each reasoner.

In the present invention, most of the 110 cross-sectional images are used, so that more accurate classification can be performed than a conventional expert diagnoses with 5 images.

In each reasoning stage, according to the symptoms of dementia, it can be classified into n stages (eg, stage 3; normal initial dementia).

3. Automatic classification reasoning

The process of diagnosing dementia using the machine-learned results according to FIG. 2 is shown in FIG. 3.

As shown, when passing through the axial plane, sagittal plane, and coronal plane inferred for a given input image, each inferring device sets a predetermined value divided into normal, early, late or normal, early, dementia stages, etc. Will be printed.

The number of each reasoner is entered into a selector, and the selector yields the most suitable result among them.

This will be described in more detail.

For the subject to be diagnosed, the sagittal and coronal images are generated from the PET image as an axial plane through a 3D model generator, and then input to the inference machine composed of the same model as the learning, and normal, early, intermediate, late or normal (normal), early It is classified into (MCI) and dementia (AD) stage (class).

In the case of deep learning, the probability for each class is output as a value between 0 and 1.

For example, for each section, values of a normal probability of 0.9, an initial probability of 0.06, and a dementia probability of 0.04 are output.

In the present invention, the final diagnosis result is determined by the selector of FIG. 3 for the inferred output value for each cross-sectional image.

,

그리고

로 정의한다.For the axial, sagittal, and coronal planes, the average values of normal, initial, and dementia classes are respectively

,

And

Is defined as

,

,

으로 정의하고, For the image of the axial plane, the output values of the deducer for normal, initial and dementia

,

,

Defined as,

,

,

, 로 정의하고, For the image of the sagittal plane, the output values of the normal, initial and dementia

,

,

Defined as,

,

,

, 로 설정하여 선택기에서 진단 결과(클래스)를 결정하는 방법은 다음과 같은 방법을 적용할 수 있다.For the coronal image, the output values of the normal, initial, dementia

,

,

The following method can be applied to the method of determining the diagnosis result (class) in the selector by setting to,.

(1) A method of determining a class with a high value by averaging the results of inferring images of each section

The selector calculates the average value for the output value for each section as shown in equations (1), (2), and (3), and calculates the diagnostic result based on the criteria shown in equation (4).

(1)

(One)

(2)

(2)

(3)

(3)

(4)

(4)

는 최종 진단결과로써 0 이면 정상, 1 이면 초기, 2 이면 치매로 분류된다. max()는 원소 중에서 최대값을 추출한다. here,

Is the final diagnosis result, and if it is 0, it is classified as normal, if it is 1, it is classified as dementia. max() extracts the maximum value from the elements.

(2) A method of determining the class with a high value based on the average of the axial, sagittal and coronal planes

The selector calculates the average value for the axial plane, the sagittal plane, and the coronal plane as in Equation (5)- Equation (7), and is determined as in Equation (8).

(5)

(5)

(6)

(6)

(7)

(7)

(8)

(8)

는 최종 진단결과로써 0 이면 정상, 1 이면 초기, 2 이면 치매로 분류된다.here,

Is the final diagnosis result, and if it is 0, it is classified as normal, if it is 1, it is classified as dementia.

(3) Method to determine the maximum value among the results of inferring images of each section

The selector determines the class with the maximum value among the outputs of the axial, sagittal, and coronal planes, that is, the class having the highest probability among the three classifications of normal, initial, and dementia as the final diagnosis result. For example, in the following cases, it is set to normal.

(9)

(9)

즉 시상면의 정상 결과가 가장 높은 값을 가지고 있기 때문에 최종 진단 결과를 정상으로 결정한 결과이다.In equation (9)

That is, since the normal result of the sagittal plane has the highest value, it is the result of determining the final diagnosis result as normal.

As described above, the present invention attempts to automatically diagnose the progression of dementia by automatically generating sagittal and coronal images based on the axial plane and then analyzing the state of the brain three-dimensionally based on all of these images.

In the case of performing the PET image analysis method based on deep learning for diagnosis and prediction of dementia according to the present invention described above, there is an advantage that a more three-dimensional and precise diagnosis of dementia is possible compared to a conventional diagnosis method based on only a specific image of the brain. .

Claims (2)

After generating a plurality of axial plane images by photographing the axial plane of the brain as tomography, image interpolation is performed on the plurality of axial plane images to generate a 3D model, and the sagittal plane of the 3D model A plurality of sagittal images and a plurality of coronal images are generated by sampling the and coronal planes at regular intervals, learning is performed using a machine learning method, and the degree of dementia is inferred using the results of the learning. As a PET image analysis method having a reasoning device for performing a dementia diagnosis that performs a process of,
Providing a predetermined axial plane image consisting of a plurality of;
Generating a plurality of predetermined sagittal images and coronal images, respectively, based on the predetermined axial plane image using a 3D model generator;
The predetermined axial plane image, the predetermined sagittal plane image, and the predetermined coronal plane image are provided to an axial plane reasoning machine, a sagittal plane reasoning machine, a coronal plane reasoning machine, and the axial plane reasoning machine, a sagittal plane reasoning machine, and Comprising the step of performing a diagnosis of dementia by inputting the result of the coronal surface inferring to a selector,
For the axial plane image, the normal, initial, and dementia output values output from the axial plane inference machine

,

,

Respectively defined as,
The normal, initial, and dementia output values output from the sagittal plane inferred for the sagittal plane image

,

,

Respectively defined as,
The normal, initial, and dementia output values output from the coronal surface inferior for the coronal surface image

,

,

If each is defined as,
The selector diagnoses the presence of dementia based on a maximum value among output values of the axial, sagittal, and coronal images. A PET image analysis method based on deep learning for dementia diagnosis and prediction, characterized in that. delete

Images