Multi-site Diagnostic Classiﬁcation Of Schizophrenia Using 3D CNN On Aggregated Task-based fMRI Data Vigneshwaran S V Bhaskaran

2025-05-02 0 0 601.92KB 6 页 10玖币

侵权投诉

Multi-site Diagnostic Classiﬁcation Of Schizophrenia Using 3D CNN On

Aggregated Task-based fMRI Data

Vigneshwaran S, V Bhaskaran

Sri Sathya Sai Institute of Higher Learning, India

{vigneshwaranpersonal@gmail.com, vbhaskaran@sssihl.edu.in }

Abstract

In spite of years of research, the mechanisms that under-

lie the development of schizophrenia, as well as its relapse,

symptomatology, and treatment, continue to be a mystery.

The absence of appropriate analytic tools to deal with the

variable and complicated nature of schizophrenia may be

one of the factors that contribute to the development of this

disorder. Deep learning is a subﬁeld of artiﬁcial intelli-

gence that was inspired by the nervous system. In recent

years, deep learning has made it easier to model and anal-

yse complicated, high-dimensional, and nonlinear systems.

Research on schizophrenia is one of the many areas of study

that has been revolutionised as a result of the outstanding

accuracy that deep learning algorithms have demonstrated

in classiﬁcation and prediction tasks. Deep learning has the

potential to become a powerful tool for understanding the

mechanisms that are at the root of schizophrenia. In addi-

tion, a growing variety of techniques aimed at improving

model interpretability and causal reasoning are contribut-

ing to this trend. Using multi-site fMRI data and a vari-

ety of deep learning approaches, this study seeks to identify

different types of schizophrenia. Our proposed method of

temporal aggregation of the 4D fMRI data outperforms ex-

isting work. In addition, this study aims to shed light on

the strength of connections between various brain areas in

schizophrenia individuals.

1 Introduction

The application of functional magnetic resonance imag-

ing (fMRI) to research human brain function and disorders

has evolved as a strong neuroimaging method. It is one of

the most crucial methods for determining how brain areas

communicate with one another to complete certain tasks.

It offers a potential method for studying the interaction be-

tween geographically distant separate brain areas that are

engaged in a task at the same time.

The blood oxygenated level-dependent (BOLD) signal in

fMRI is not random, but rather temporally coherent between

geographically distant functionally consistent areas. Func-

tional brain networks are formed by functionally connected

areas (Biswal, Yetkin, Haughton, & Hyde, 1995; Cordes

et al., 2000). The identiﬁcation of functional activation in

fMRI data during task or task-free ”resting state” is crucial

in neuroscience for pre-surgical planning and the diagnosis

of neuropsychiatric diseases such as Alzheimer’s (Jones et

al., 2012), autism (Starck et al., 2013), and schizophrenia

(Sako˘

glu et al., 2010).

Several ways have been tested to diagnose diseases or

classify persons as normal or ill. In general, the low SNR of

fMRI data presents difﬁculties in using this data to diagnose

neuropsychiatric disorders. To the best of our knowledge,

no standard automated fMRI instrument is available in hos-

pitals for illness diagnosis. This has prompted researchers

to investigate the development of such systems that can as-

sist doctors. Researchers, for example, have used station-

ary functional connectivity (FC) as a biomarker to differ-

entiate patients with neurological and psychiatric diseases

such as Alzheimer’s (Huang et al., 2010) or Schizophrenia

from normal subjects (Fox & Greicius, 2010) because FC

has been discovered to be changed in several neuropsycho-

logical diseases (Jones et al., 2012; Leonardi et al., 2013;

Leonardi & Van De Ville, 2013; X. Li et al., 2014).

In this study, we ﬁrst collected fMRI raw datasets from

two publicly available repositories including 300+ partici-

pants and developed a 3D convolutional network to train on

the images by aggregating information over the timesteps

for every subject for the automatic diagnosis of individuals

with schizophrenia. We compare our model with various

baseline models and record the performance.

2 Related Work

Machine learning has had a signiﬁcant impact on

schizophrenia diagnosis studies. Along with MRI and

fMRI, the researchers analysed data from genetics, elec-

troencephalography, and even audio interviews (Cortes-

arXiv:2210.05240v1 [cs.CV] 11 Oct 2022

Briones, Tapia-Rivas, D’Souza, & Estevez, 2021). A com-

bination of fMRI and genomics data from a single loca-

tion was utilized to perform canonical correlation analy-

sis using two fully linked, sparse autoencoders followed by

SVM (G. Li et al., 2020). There were also several stud-

ies that combined fMRI and structural MRI; the fMRI tasks

included resting state, letter-n back task, auditory oddball,

and audiovisual stimuli tasks that elicited negative and neu-

tral emotion (Dakka et al., 2017; Kim, Calhoun, Shim, &

Lee, 2016; Oh et al., 2019; Salvador et al., 2019; Yan et al.,

2019). These studies used a variety of strategies to differen-

tiate between classes, including 3D activation maps, func-

tional connectomes, ICA decomposition, and independent

polygenic risk scores along with modelling techniques such

as artiﬁcial neural networks and decision tree classiﬁers.

Our research was inspired by a publication in which the

authors attempted to discover consistent patterns and con-

nections between brain areas and also categorized partici-

pants using a variety of machine learning models (Gheirat-

mand et al., 2017). However, their study was focused on

a single site, and we desired to incorporate data from nu-

merous locations, which is where (Zeng et al., 2018) came

in handy. The writers of this article combined private and

publicly available datasets. That, we believe, is the only ef-

fort in this area that utilized such a vast sample space for

model training. We were unable to locate many open fMRI

datasets, particularly those that included both T1 weighted

and resting-state fMRI. Finally, we chose two locations,

COBRE and UCLA.

3 Materials and Methods

3.1 Participants

The dataset is made up of 120 schizophrenia patients and

206 healthy people who were gathered from two different

imaging resources. Refer to Table 1. The ﬁrst subgroup

was developed at the University of California, Los Ange-

les (UCLA: 50 schizophrenics and 122 healthy controls;

patients were allowed to use stable medications) (Bilder

et al., 2016). The second group comes from the Center

for Biomedical Research Excellence (COBRE) (COBRE:

69 Schizophrenia Strict Spectrum patients, 11 Schizoaffec-

tive Spectrum patients, and 84 healthy controls; all of the

patients were on antipsychotic medications) (Aine et al.,

2017).

All individuals were screened and rejected if they had

a history of neurological illness, mental retardation, seri-

ous head trauma resulting in a loss of consciousness lasting

more than 5 minutes, or a history of substance addiction

or dependency within the previous 12 months. The Struc-

tured Clinical Interview for DSM-IV Disorders was utilized

to obtain diagnostic information (SCID).

3.2 Image Acquisition

On a Siemens Erlangen 3.0 Tesla Trim Trio scanner, MR

images from both locations were collected. There were,

however, modest changes in the methods used to capture

the photos. The COBRE dataset was collected in complete

k-space EPI sequences in a single shot. The time repetitions

were set to 2000 milliseconds, and the echo time was set to

29 milliseconds. With a ﬁeld of vision of 192 mm and a

thickness of 4 mm, there is no gap between the 32 slices. A

resting state run was performed on each participant, yield-

ing 150-time steps. An asymmetrical spin-echo echo-planar

sequence was used to collect the UCLA dataset. The time

repetitions were set to 2000 milliseconds, and the echo time

was set to 30 milliseconds. With a ﬁeld of vision of 192

mm and a thickness of 4 mm, there is no gap between the

34 slices. A resting state run was performed on each partic-

ipant, yielding 152-time steps.

3.3 Data Preprocessing

We performed an extensive 7-step pipeline to preprocess

the resting-state fMRI data. It has been recorded by re-

search that enhanced preprocessing of the fMRI data im-

proves the performance of modelling (Churchill, Spring,

Afshin-Pour, Dong, & Strother, 2015). First, we apply

slice-time correction to each voxel’s time series using SPM

known as Hanning-Windowed Sinc Interpolation (HWSI).

Afterwards, we subtract 7 timesteps from the beginning of

each subject to account for magnetic saturation. Using FSL-

MCFLIRT (Jenkinson, Bannister, Brady, & Smith, 2002),

we were able to correct the fMRI for motion artefacts in-

duced by low-frequency drifts, which might have a detri-

mental inﬂuence on the time course decomposition of the

data. We eliminated the skull and neck voxels from the

structural T1 weighted image using the FSL-BET (Smith,

2002) corresponding to each fMRI time series from the

structural T1 weighted image. As a result, we coregistered

the structural image with the functional image, and the pro-

cess was completed. Following that, we matched the regis-

tered brains to the Montreal Neurological Institute standard

3mm brain template (MNI152) (Fonov et al., 2011). Us-

ing a Gaussian kernel with a full-width half-maximum of

4mm, spatial smoothing was performed to a time series of

data (FWHM). The ﬁnal step was to downscale the images

to the integer format in order to decrease the amount of disc

space consumed.

3.4 Functional Connectivity Measure

In order to extract the functional connectomes, we used

two brain atlases for two different purposes. One was to

use it for classiﬁcation purposes whereas the other one is

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：

Multi-siteDiagnosticClassicationOfSchizophreniaUsing3DCNNOnAggregatedTask-basedfMRIDataVigneshwaranS,VBhaskaranSriSathyaSaiInstituteofHigherLearning,Indiafvigneshwaranpersonal@gmail.com,vbhaskaran@sssihl.edu.ingAbstractInspiteofyearsofresearch,themechanismsthatunder-liethedevelopmentofschizophrenia...

展开>> 收起<<

Multi-site Diagnostic Classiﬁcation Of Schizophrenia Using 3D CNN On Aggregated Task-based fMRI Data Vigneshwaran S V Bhaskaran.pdf

共6页,预览2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！

Multi-site Diagnostic Classiﬁcation Of Schizophrenia Using 3D CNN On Aggregated Task-based fMRI Data Vigneshwaran S V Bhaskaran

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: