DEEPMULTI -BRANCH CNN A RCHITECTURE FOR EARLY ALZHEIMER SDETECTION FROM BRAIN MRI S Paul K. Mandal

2025-05-06 0 0 1.57MB 11 页 10玖币

侵权投诉

DEEP MULTI-BRANCH CNN ARCHITECTURE FOR EARLY

ALZHEIMER’SDETECTION FROM BRAIN MRIS

Paul K. Mandal*

Department of Computer Science

University of Texas at Austin

Austin, TX 78712 USA

mandal(at)utexas.edu

*Corresponding author

Rakeshkumar Mahto

Department of Electrical and Computer Engineering

California State University Fullerton

Fullerton, CA 92831 USA

ramahto(at)fullerton.edu

for the Alzheimer’s Disease Neuroimaging Initiative∗

June 21, 2023

ABSTRACT

Alzheimer’s disease (AD) is a neuro-degenerative disease that can cause dementia and result severe

reduction in brain function inhibiting simple tasks especially if no preventative care is taken. Over

1 in 9 Americans suffer from AD induced dementia and unpaid care for people with AD related

dementia is valued at $271.6 billion. Hence, various approaches have been developed for early AD

diagnosis to prevent its further progression. In this paper, we ﬁrst review other approaches that

could be used for early detection of AD. We then give an overview of our dataset that was from

the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and propose a deep Convolutional Neural

Network (CNN) architecture consisting of 7,866,819 parameters. This model has three different

convolutional branches with each having a different length. Each branch is comprised of different

kernel sizes. This model can predict whether a patient is non-demented, mild-demented, or moderately

demented with a 99.05% three class accuracy.

Keywords Alzheimer’s, Brain Imaging, CNN, Convolution, Convolutional Neural Network, Deep Learning, Disease

Detection, Neural Network, Machine Learning, Medical Diagnosis

1 Introduction

Alzheimer’s Disease (AD) is a common disease that affects 1 in 9 (10.7%) Americans over 65. 6.5 million Americans

aged 65 or over have been diagnosed with AD dementia. An estimated 16 billion unpaid man-hours of care were given

to people with dementia from AD in 2021 which has an estimated value of $271.6 billion[1]. Approximately 12% to

18% of people over 60 are living with mild cognitive impairment (MCI)[2]. MCI causes subtle changes in memory

and thinking. Although often associated with the normal aging process, MCI is not apart of typical aging. Moreover,

10%-15% of individuals with MCI develop full dementia each year[3]. Therefore, AD must be diagnosed at an early

stage to prevent it from progressing further. For this purpose, Machine Learning (ML) and Deep Learning (DL) can

play an invaluable role since they have been extensively used in various other medical applications for diagnosing and

detecting various abnormalities and diseases[4–6].

∗

Data used in preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database

(adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or

provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found at:

http://adni.loni.usc.edu/wp-content/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf

arXiv:2210.12331v3 [eess.IV] 17 Jun 2023

APREPRINT - JUNE 21, 2023

A diverse range of approaches has been employed in the ﬁeld of early Alzheimer’s disease (AD) detection, encompassing

the analysis of speech patterns and inﬂections, neuropsychometric tests, olfactory tests, eye testing, gait testing, as well

as the utilization of neural networks on various diagnostic modalities such as MRIs, electroencephalograms (EEG), and

magnetoencephalographs (MEG) [7]. Recently, there has been a surge in the popularity of machine learning and deep

learning techniques for early AD diagnosis, with a predominant focus on applying these methods to MRI images, MEG,

EEG, and other relevant physiological parameters[8].

1.0.1 Neural networks on MRIs

Over the years, various neural network techniques have been used to predict Alzheimer’s. In [9], a convolution neural

network (CNN) was applied to accurately predict mild cognitive impairment to AD. The overall accuracy reported in

[9] was around 86.1%. A similar technique of CNN was applied to a dataset consisting of MRI images of 156 AD and

156 normal patients [10]. The dataset in this study consisted of AD patients and age/gender-matched normal individuals

[10]. The proposed technique in [10] achieved an accuracy of 94%. A mix of 2D CNN and recurrent neural networks

(RNN) on MRI images were reported to achieve an accuracy of 96.88% [11]. The proposed technique applied an

RNN after applying a 2D CNN to recognize the connection between 2D image slices [11]. The study also presented a

technique of transfer learning from 2D images to 3D CNNs. One of the best-performing models that didn’t rely on

MRIs was a neural network trained to analyze speech patterns. One of their models reported a 97.18% accuracy [12].

However, there are two main issues with this approach. First and foremost is that it is clear from looking at the audio

waves that the subjects who have AD are well past the MCI/Mild Demented stage making it non-viable for an early

detection stage. The second is that the study only had 50 non-demented subjects and 20 demented subjects. Each

non-demented subject had 12 hours of audio and each demented subject had 8 hours of audio. These clips were divided

into 600 different clips of 60 second audio. However, the paper does not state whether they divided the training and

validation sets by patient or not. If that is the case, there is a possibility that the neural network is learning how to

classify whether the subject has AD based on the patient’s voice, rather than extracting useful information.

1.0.2 Neural networks on magnetoencephalographs (MEG)

Compared to MRI images, a non-invasive diagnostic technique called Magnetoencephalography (MEG) is utilized

for measuring brain activity. Based on brain activity, the proposed method estimates the magnetic ﬁeld generated by

the slow ionic current ﬂow through cells. This research shows that MEG activity can provide excellent sensitivity for

early diagnosis of DP [13]. A combination of the MEG recording and MRI scans are utilized in [14], which resulted

in an accuracy of 89%. A similar technique for diagnosing AD was presented in [15]. However, the accuracy of the

classiﬁcation technique was 77%. Various other machine learning (ML) driven techniques for diagnosing AD using

MEGs are summarized in [16]. However, none of the techniques were able to achieve an accuracy greater than than

90%.

1.0.3 Neural networks on electroencephalograms (EEG)

Another more promising study in AD diagnosis has been done with EEGs. Electrophysiological imaging techniques

such as EEGs are widely accepted as reliable indicators for the diagnosis of AD. With the aid of neural networks, it has

become possible to use EEG data to accurately determine whether a patient has Alzheimer’s disease. A novel neural

network, I-Fast, was able to predict whether subjects had AD with 92% accuracy [17]. The dataset used in this study

consisted of 115 mild cognitive impairment and 180 AD patients. This is signiﬁcant as it implies that EEGs can be

used as a viable alternative for the diagnosis of AD, given the cost effectiveness of the technique. Similarly, a novel

technique was presented in [18] that uses a ﬁnite response ﬁlter (FIR) in a double time domain to extract features from

an EEG recording dataset consisting of MCI, AD, and healthy controls (HC). Later, Binary Classiﬁcation (BC) achieved

an accuracy of 97%, 95%, and 83% between HC vs. AD, HC vs. MCI, and MCI vs. AD, respectively.

1.0.4 Blood Plasma

Another approach was to test for a panel of 18 different proteins from blood samples. This approach was able to achieve

an 89% accuracy[19]. This is probably the most promising of the methods described above since it is much easier and

less costly to run blood tests. Although we do concede that there are beneﬁts to the techniques outlined above due to the

limited availability of MRIs, none of the approaches enumerated above were able to exceed our 99% accuracy achieved

from our approach.

APREPRINT - JUNE 21, 2023

Figure 1: A sample of 9 preprocessed images from the ADNI dataset.

1.1 Description of Alzheimer’s MRI Datasets

Our dataset consists of 6338 magnetic resonance imaging (MRI) images that were imaged from the Alzheimer’s Disease

Neuroimaging Initiative (ADNI)[20] and were curated and preprocessed on Kaggle[21]. Our preprocessed dataset came

formatted in 100x100 pixel images. Our dataset consists of 3202 images of non-demented patients, 2242 images of very

mild demented patients, and 892 images of mildly demented patients as we opted to not use the 64 moderate demented

images due to the low sample size. Figure 1 shows 9 images from the ADNI dataset.

We divided our training and test set into 5701 training images (2881 non-demented, 2017 very mild demented, and 802

mild demented) and 637 test images (321 non-demented, 225 very mild demented, and 90 mild demented) as shown

in Figure 2. We used stratiﬁed random sub-sampling to ensure that the training and test sets had the same ratio of

non-demented, very mild demented, and mild demented images.

Figure 2: Distribution of Training and Test images from the ADNI dataset.

2 Background

Signiﬁcant gains have been made in image recognition and object detection through the use of deep learning. These

advances have been applied to medical imaging such as diabetic retinapathy [22]. For the purposes of this paper,

familiarity of the subsequent concepts is necessary.

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

DEEPMULTI-BRANCHCNNARCHITECTUREFOREARLYALZHEIMER’SDETECTIONFROMBRAINMRISPaulK.Mandal*DepartmentofComputerScienceUniversityofTexasatAustinAustin,TX78712USAmandal(at)utexas.edu*CorrespondingauthorRakeshkumarMahtoDepartmentofElectricalandComputerEngineeringCaliforniaStateUniversityFullertonFullerton,CA...

展开>> 收起<<

DEEPMULTI -BRANCH CNN A RCHITECTURE FOR EARLY ALZHEIMER SDETECTION FROM BRAIN MRI S Paul K. Mandal.pdf

共11页,预览3页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

DEEPMULTI -BRANCH CNN A RCHITECTURE FOR EARLY ALZHEIMER SDETECTION FROM BRAIN MRI S Paul K. Mandal

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: