Deep Learning Type Convolution Neural Network Architecture for

Multiclass Classification of Alzheimer’s Disease

Gopi Battineni

, Nalini Chintalapudi

, Francesco Amenta

and Enea Traini

Telemedicine and Tele Pharmacy Centre, School of Medicinal and Health Products Sciences,

University of Camerino, Camerino, 62032, Italy

Keywords: Alzheimer’s Disease (AD), OASIS-3, MRI Images, Deep Learning, CNN.

Abstract: Alzheimer’s disease (AD) is one of the common medical issues that the world is facing today. This disease

has a high prevalence of memory loss and cognitive decline primarily in the elderly. At present, there is no

specific treatment for this disease, but it is thought that identification of it at an early stage can help to manage

it in a better way. Several studies used machine learning (ML) approaches for AD diagnosis and classification.

In this study, we considered the Open Access Series of Imaging Studies-3 (OASIS-3) dataset with 2,168

Magnetic Resonance Imaging (MRI) images of patients with very mild to different stages of cognitive decline.

We applied deep learning-based convolution neural networks (CNN) which are well-known approaches for

diagnosis-based studies. The model training was done by 70% of images and applied 10-fold cross-validation

to validate the model. The developed architecture model has successfully classified the different stages of

dementia images and achieved 83.3% accuracy which is higher than other traditional classification techniques

like support vectors and logistic regression.

INTRODUCTION

Alzheimer's Disease (AD) is the most well-known

and largely diffused neurodegenerative disorder

occurring in the elderly. AD negatively affects

patients' everyday lives, causing an advanced decline

of cognitive capabilities such as memory, language,

behaviour, and critical thinking (Alzheimer’s Disease

International (ADI ) 2010). Changes in cognitive

impairment of AD patients start slowly and evolve

rapidly over the long run.

Similar to other body parts, brain can change as

people get older. Some people lost thinking and

incidental issues with recollecting certain things.

Excessive cognitive decline, and other significant

changes in the manner in which brain function is

impaired (Jaussent et al. 2012). The first symptoms of

AD are trouble recalling recently learned data

because Alzheimer's progressions regularly start in

the brain areas involved in learning and memory. As

Alzheimer's progresses progressively severe

symptoms like confusion, mood changes,

disorientation, unwarranted doubts about family and

companions, and trouble talking appear. Individuals

with cognitive decline or other potential indications

of AD may think that it’s difficult to remember they

have an issue.

AD is a type of dementia with several

implications on the cognitive domain, affecting

primarily thinking and memory. Specialists and

different parental figures screen the movement of AD

in patients by assessing the level of decrease in the

patients' psychological capacities that are often

classified into three stages: very mild (normal

cognitive), mild cognitive impairment (MCI), and

demented (Gaugler et al. 2016). Figure 1 presents

the magnetic resonance image (MRI) images of

different AD conditions. Although the MCI and

dementia patients both are experiencing a reduction

of cognitive abilities, dementia patients would suffer

from more pronounced difficulties with thinking or

hampered judgment.

https://orcid.org/0000-0003-0603-2356

https://orcid.org/0000-0003-0818-306X

https://orcid.org/0000-0002-0555-1034

Battineni, G., Chintalapudi, N., Amenta, F. and Traini, E.

Deep Learning Type Convolution Neural Network Architecture for Multiclass Classiﬁcation of Alzheimer’s Disease.

DOI: 10.5220/0010378602090215

In Proceedings of the 14th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2021) - Volume 2: BIOIMAGING, pages 209-215

ISBN: 978-989-758-490-9

209

Figure 1: AD presented by MRI images (a) mild dementia;

(b) moderate demented; (c) nondemented; and (d) very

mild demented.

In clinical practice, the capacity to accurately

forecast the patient diagnosis can help by adding

appropriate medical decisions on treatment

approaches. Recently, machine learning (ML)

algorithms are largely applying to forecast and predict

diseases and helping in quick decision making

(Battineni, Sagaro, et al., 2020). Pattern-related

approaches like logistic regression (Johnson et al.,

2014), support vector machines (Battineni,

Chintalapudi, en Amenta 2019), and linear

discriminant analysis (Rathore et al. 2017) are

giving promising results in the prediction of AD

development and early AD detection.

Deep learning models were used unlabeled data

during preprocessing. These are well suited for

imbalanced datasets and achieve a knowledge base

(Mittal et al. 2019). At present these are largely

involved in all other problems that are not able to be

addressed by traditional artificial intelligence (AI)

techniques. Neural networks are the latest deep

learning algorithms that have discovered the

functionality of different situations. Convolutional

neural networks (CNN) are characterized

contributions to profits through a complex

composition of layers that presents building blocks

including nonlinear functions and transformations.

Medical experts feel that deep learning could be a

promising solution in AD identification and stage

detection (Khan et al., 2020). For instance,

(Basheera en Sai Ram, 2019) applied CNN

modeling for AD diagnosis based on T2 weighted

magnetic resonance imaging (MRI) and achieved

90.47% accuracy. A Siamese CNN can also help to

categorize the AD and studies reported 99.05% of

accuracy (Mehmood et al. 2020). It is also reported

that AD prediction from MCI using the CNN model

reported 79.9% of accuracy(Lin et al., 2018).

Therefore, it is assumed that an effective and

comprehensive deep learning model can help to

identify early AD prediction and ultimately provide

timely treatment to

the suffered patients. In this work, we proposed

convolutional neural networks (CNN) model of

deep learning type for detection of early-stage AD

and successfully classify the MRI images on four

different dementia stages presented in Figure 2.

Experiments were conducted on longitudinal

neuroimages of the OASIS-3 database that include

MR scans of T1-weighted, T2 weighted, ASL, SWI,

DTI sequences, FLAIR, time of flight, and resting-

state BOLD. The rest of the paper is structured

according to the following outline: Section 2 presents

the dataset and proposed model architecture; section

3 presents the experimental results, and section 4

makes a discussion which is followed by the

conclusion in section 5.

Figure 2: Brain image classification with the CNN model

framework.

METHODS

2.1 Dataset

The Open Access Series of Imaging Studies

(OASIS) contains MR scanning information that is

openly accessible to scientific communities. They

released OASIS-1 (cross-sectional) and OASIS-2

(longitudinal) MRI datasets among different subjects

and these datasets are widely used in many studies

(Sweeney et al. 2013; Palumbo et al. 2019). OASIS-

3 is the extension of previous datasets. It includes

1,098 patients aging from 42 to 95 years. Among

participants, 609 are associated with normal cognitive

decline (very mild), and 489 were associated with

different cognitive decline stages. OASIS-3 dataset

incorporated both functional and structural features of

more than 2,000 MRI images. The dataset outcome of

four categories of MR images has presented in Figure

BIOIMAGING 2021 - 8th International Conference on Bioimaging

210

Figure 3: Dataset outcome of different dementia stages

(3*4 image matrix).

2.2 CNN Model Architecture

A convolutional neural network (ConvNet) is deep

learning type algorithms that take images as input,

assign features based on their importance (biases and

learnable weights) to different image objects, and

also be able to separate one from the other

(Krizhevsky, Sutskever, en Hinton 2017). When

compared with other classification models, ConvNet

possesses low complex pre-processing steps. In CNN,

each input image is gone through sequence

convolution layers namely pooling layers, filtering

layers (kernels), and fully connected layers (FCs).

To make the proposed model easier for

understanding, we created a dense layer block and

convolution block. The architecture of the CNN

model is inspired by the article (Pan et al. 2020). We

built the CNN model by using five convolutional

slabs covered with convolution layers, feature

engineering, max pooling, and classification. We

have used cross-entropy as a loss function and Adam

as an optimizer. SoftMax has been used to classify the

multiclass AD stages since it is associated with a

mutually exclusive relationship. The feature

representation (f

) works as an input to the SoftMax

layer and interprets output brain stages. A probability

score P (k) for each class as defined as

 

∑









;where fi feature representation, and

Cross entropy loss function as

(L)=

∑

𝑡𝑘. log 𝑝𝑘





; where t

ground truth of

MRimage then





-t

2.3 Experimental Setup

Figure 4 presents the most relevant procedures

followed to construct the feature data of brain images

and extraction of AD images developed in this paper.

After pre-processing steps, the given image dataset

has been divided into training and validation files

with standard (80:20) division.

The procedures indicated red line are MR images

that fed to the CNN model for training purposes.

The model extracts the input image features of

trained images under present parameters and supplies

them to the SoftMax classifier for testing. The

SoftMax function calculates the loss and model

accuracy. For avoiding high loss, network

parameters are adjusted by the back-propagation

algorithm. After applying several iterations (epochs)

the better-trained parameters have been achieved.

The model visualization metrics like loss and

receiver operating characteristic area under the curve

(ROC AUC) have been taken as the performance

parameter for AD classification since it has been

considered one of the key metrics in multi-image

classification techniques. The experimental setup

and AD detection and classification have been done

through TensorFlow and python language.

Deep Learning Type Convolution Neural Network Architecture for Multiclass Classiﬁcation of Alzheimer’s Disease

211

Figure 4: Experimental setup of the work.

RESULTS

To do efficient training on our CNN model, a back-

propagation algorithm is set to adjust the rate of

learning and stop the model automatically once it

reaches maximum accuracy. Since the learning rate is

one of the hyperparameters that decides model

accuracy and time to process the model. OASIS-3

dataset consisted of 2168 independent MRI

scanners. Among the given images, 1,734 are used

for training and 434 were used for validation

purposes. Because of the large image dataset, 10-

fold cross-validation has been used and we have

used each fold 70% as training, 10% as validation,

and 20% images are used testing. The distribution of

the dataset is presented in Table 1.

Table 1: Total image distribution.

Total Images: 2168

Type Percentage

Trained images 1517 (70%)

Testing images 434 (20%)

Validation images 217 (10%)

The model-fitting has to be done on a sample of

100 epochs and to prevent model overfitting we stop

the model early at the 80

iteration. The model took

a run time of 138 min to process the trained images.

Figure 5 presents a graphical representation of ROC.

AUC and loss metrics after each iteration on both

training and validation image data.

Figure 5: Model AUC and loss metric outcomes.

Though the model evaluation has been done on

the validation dataset, we also perform the

BIOIMAGING 2021 - 8th International Conference on Bioimaging

212

experiments on the testing dataset. The testing

dataset model AUC curve outcome has presented in

Figure 6 and the model achieved a ROC of 83.3%

which is considered as an optimal classifier for AD

image detection and this value is significantly higher

than traditional ML approaches (Battineni,

Chintalapudi, en Amenta 2019; A. Khan en Zubair

2020).

Figure 6: The ROC curve of test data.

DISCUSSION

In this work, we presented a novel deep learning

type CNN model for the classification of AD subjects.

As mentioned above, AD is the most common adult-

onset dementia and contributes about 60-70% of

worldwide dementia cases (A. Khan en Zubair 2020).

Unfortunately, there is no proper medication or cure

for AD, and advancements in AD cure have been

getting slow. Screening among people of AD risk

given electronic health records (EHR) in preclinical

stages may prompt early identification of AD

pathology and suggest better approaches for

complying with the AD beginning. Current

biomarkers of AD have required specimen collection

(like serum or liquid), MRI image data, or more

sophisticated markers that at the present can be

identified just in highly specialized centres

(Mantzavinos en Alexiou 2017; Hadjichrysanthou et

al. 2020).

On the other hand, the EHRs for example medical

records in clinical settings, or administrative health

information don't require extra time or effort for data

collection. Likewise, with the coming of

digitalization, the measures of such information have

drastically increased (Shao et al. 2019). Since it is

omnipresent, enormous, and cost-effective, the

digitized medical database might be a significant

asset for testing different AD predictive models.

Nonetheless, despite its enormous possible value,

somehow thought about the degrees to which the

enormous scope of EHR data can help in risk of AD

prediction (Shao et al. 2019; Mayer et al. 2015). The

possible prediction of future AD progression is

incredibly significant in clinical practice also, in

healthcare research. Advanced neuroimaging

techniques like MRI, positron emission tomography

(PET) is developed and presented to identify AD-

related molecular and structural biomarkers

(Hadjichrysanthou et al. 2020).

Computer scientists are recommending applying

sophisticated computing techniques like machine

learning and deep learning. It is reported that 99.1%

of accuracy has been achieved through the application

of ensemble learning models for late-life AD

detection among 150 patients (Battineni,

Chintalapudi, et al. 2020). AD prediction among 123

subjects with Pre-MCI and MCI was done by

clinically transmittable ML algorithms and results

reported the whole sample accuracy of 96.2% (Grassi

et al. 2018). However, most of the outcomes proposed

by these algorithms are based on demographic

magnetic resonance image (MRI) information.

Because of this, researchers believed that deep

learning algorithms are the best approaches if brain

images were included (Choi en Jin 2018). Most of the

works associated with Machine learning in the early

prediction of AD occurred with high success. For

instance, it is reported that 94.1% of accuracy by 3D

convolutional neural networks (CNN)

(Esmaeilzadeh et al. 2018).

This work presented a deep CNN with 10-fold

cross-validation and achieved more than 80%

accuracy. While applying computing methods for

diagnosis, a small portion of datasets are presented.

Therefore, our model maintained a random image

selection of train, test, and validation datasets. The

proposed model produced promising results in AD

image classification. The most notable outcome for

this study is the progressions among predictiveness of

AD diseases.

CONCLUSIONS

An autonomous AD detection classifier based deep

ConvNet framework is presented. We adopted the

latest release of the OASIS-3 dataset that contains

Deep Learning Type Convolution Neural Network Architecture for Multiclass Classiﬁcation of Alzheimer’s Disease

213

different categories of AD datasets. For training,

more than 1,500 images model took a bit longer

process than expected, but it is faster than mankind

process. Deep ConvNets do not need any handcrafted

feature selection approach because of having

autonomous feature tuning. The main limitation of

the study is to adopt only a single classifier for the

brain MRI data classification and there are other

possibilities to do better improvements in the

proposed model architecture. Although attained

results of higher 80% accuracy while compared over

traditional ML classifiers, many advancements are

proposed to enhance the model quality.

CONFLICTS OF INTEREST

No author has produced any conflicts of interest.

REFERENCES

Alzheimer’s Disease International (ADI). 2010. “World

Alzheimer Report 2010: The Global Economic Impact

of Dementia”. Alzheimer’s Disease International (ADI).

https://doi.org/10.1111/j.0963-7214.2004.00293.x.

Basheera, Shaik, en M. Satya Sai Ram. 2019. “Convolution

neural network–based Alzheimer’s disease

classification using hybrid enhanced independent

component analysis based segmented gray matter of

T2 weighted magnetic resonance imaging with clinical

valuation”. Alzheimer’s and Dementia: Translational

Research and Clinical Interventions. https://doi.org/

10.1016/j.trci.2019.10.001.

Battineni, Gopi, Nalini Chintalapudi, en Francesco Amenta.

2019. “Machine learning in medicine: Performance

calculation of dementia prediction by support vector

machines (SVM)”. Informatics in Medicine

Unlocked. https://doi.org/10.1016/j.imu.2019.100200.

Battineni, Gopi, Nalini Chintalapudi, Francesco Amenta, en

Enea Traini. 2020. “A Comprehensive Machine-

Learning Model Applied to Magnetic Resonance

Imaging (MRI) to Predict Alzheimer’s Disease (AD) in

Older Subjects”. Journal of Clinical Medicine.

https://doi.org/10.3390/jcm9072146.

Battineni, Gopi, Getu Gamo Sagaro, Nalini Chinatalapudi,

en Francesco Amenta. 2020. “Applications of machine

learning predictive models in the chronic disease

diagnosis”. Journal of Personalized Medicine.

https://doi.org/10.3390/jpm10020021.

Choi, Hongyoon, en Kyong Hwan Jin. 2018. “Predicting

cognitive decline with deep learning of brain

metabolism and amyloid imaging”. Behavioural Brain

Research. https://doi.org/10.1016/j.bbr.2018.02.017.

Esmaeilzadeh, Soheil, Dimitrios Ioannis Belivanis, Kilian

M. Pohl, en Ehsan Adeli. 2018. “End-to-end

alzheimer’s disease diagnosis and biomarker

identification”. In Lecture Notes in Computer Science

(including subseries Lecture Notes in Artificial

Intelligence and Lecture Notes in Bioinformatics).

https://doi.org/10.1007/978-3-030- 00919-9_39.

Gaugler, Joseph, Bryan James, Tricia Johnson, Ken Scholz,

en Jennifer Weuve. 2016. “2016 Alzheimer’s disease

facts and figures”. Alzheimer’s and Dementia.

https://doi.org/10.1016/j.jalz.2016.03.001.

Grassi, Massimiliano, Giampaolo Perna, Daniela Caldirola,

Koen Schruers, Ranjan Duara, en David A.

Loewenstein. 2018. “A clinically-translatable machine

learning algorithm for the prediction of Alzheimer’s

disease conversion in individuals with mild and

premild cognitive impairment”. Journal of Alzheimer’s

Disease. https://doi.org/10.3233/JAD- 170547.

Hadjichrysanthou, Christoforos, Stephanie Evans, Sumali

Bajaj, Loizos C. Siakallis, Kevin McRae-Mckee, en

Frank De Wolf. 2020. “The dynamics of biomarkers

across the clinical spectrum of Alzheimer’s disease”.

Alzheimer’s Research and Therapy. https://doi.org/

10.1186/s13195-020-00636-z.

Jaussent, Isabelle, Jean Bouyer, Marie Laure Ancelin,

Claudine Berr, Alexandra Foubert-Samier, Karen

Ritchie, Maurice M. Ohayon, Alain Besset, en Yves

Dauvilliers. 2012. “Excessive sleepiness is predictive

of cognitive decline in the elderly”. Sleep.

https://doi.org/10.5665/sleep.2070.

Johnson, Piers, Luke Vandewater, William Wilson, Paul

Maruff, Greg Savage, Petra Graham, Lance S.

Macaulay, et al. 2014. “Genetic algorithm with logistic

regression for prediction of progression to Alzheimer’s

disease”. BMC Bioinformatics. https://doi.org/10.

1186/1471-2105-15-S16-S11.

Khan, Afreen, en Swaleha Zubair. 2020. “An Improved

Multi-Modal based Machine Learning Approach for

the Prognosis of Alzheimer’s disease”. Journal of

King Saud University - Computer and Information

Sciences. https://doi.org/10.1016/j.jksuci.2020.04.004.

Khan, Mehshan Ahmed, Muhammad Attique Khan,

Fawad Ahmed, Mamta Mittal, Lalit Mohan Goyal, D.

Jude Hemanth, en Suresh Chandra Satapathy. 2020.

“Gastrointestinal diseases segmentation and

classification based on duo-deep architectures”.

Pattern Recognition Letters. https://doi.org/10.1016/

j.patrec.2019.12.024.

Krizhevsky, Alex, Ilya Sutskever, en Geoffrey E. Hinton.

2017. “ImageNet classification with deep

convolutional neural networks”. Communications of

the ACM. https://doi.org/10.1145/3065386.

Lin, Weiming, Tong Tong, Qinquan Gao, Di Guo,

Xiaofeng Du, Yonggui Yang, Gang Guo, Min Xiao,

Min Du, en Xiaobo Qu. 2018. “Convolutional neural

networks-based MRI image analysis for the

Alzheimer’s disease prediction from mild cognitive

impairment”. Frontiers in Neuroscience. https://doi.

org/10.3389/fnins.2018.00777.

Mantzavinos, Vasileios, en Athanasios Alexiou. 2017.

“Biomarkers for Alzheimer’s Disease Diagnosis”.

BIOIMAGING 2021 - 8th International Conference on Bioimaging

214

Current Alzheimer Research. https://doi.org/10.2174/

15672050146661702031259 42.

Mayer, Miguel A., Laura I. Furlong, Pilar Torre, Ignasi

Planas, Francesc Cots, Elisabet Izquierdo, Jordi

Portabella, Javier Rovira, Alba Gutierrez-Sacristan, en

Ferran Sanz. 2015. “Reuse of EHRs to Support

Clinical Research in a Hospital of Reference”. In

Studies in Health Technology and Informatics.

https://doi.org/10.3233/978-1-61499-512-8-224.

Mehmood, Atif, Muazzam Maqsood, Muzaffar Bashir, en

Yang Shuyuan. 2020. “A deep siamese convolution

neural network for multi-class classification of

alzheimer disease”. Brain Sciences. https://doi.org/

10.3390/brainsci10020084.

Mittal, Mamta, Lalit Mohan Goyal, Sumit Kaur, Iqbaldeep

Kaur, Amit Verma, en D. Jude Hemanth. 2019. “Deep

learning based enhanced tumor segmentation approach

for MR brain images”. Applied Soft Computing

Journal. https://doi.org/10.1016/j.asoc.2 019.02.036.

Palumbo, L., P. Bosco, M. E. Fantacci, E. Ferrari, P. Oliva,

G. Spera, en A. Retico. 2019. “Evaluation of the intra-

and inter-method agreement of brain MRI

segmentation software packages: A comparison

between SPM12 and FreeSurfer v6.0”. Physica

Medica. https://doi.org/10.1016/j.ejmp.2019.07.016.

Pan, Dan, An Zeng, Longfei Jia, Yin Huang, Tory Frizzell,

en Xiaowei Song. 2020. “Early Detection of

Alzheimer’s Disease Using Magnetic Resonance

Imaging: A Novel Approach Combining

Convolutional Neural Networks and Ensemble

Learning”. Frontiers in Neuroscience. https://doi.org/

10.3389/fnins.2020.00259.

Rathore, Saima, Mohamad Habes, Muhammad Aksam

Iftikhar, Amanda Shacklett, en Christos Davatzikos.

2017. “A review on neuroimaging-based classification

studies and associated feature extraction methods for

Alzheimer’s disease and its prodromal stages”.

NeuroImage. https://doi.org/10.1016/j.neuroimage.

2017.03.057.

Shao, Yijun, Qing T. Zeng, Kathryn K. Chen, Andrew

Shutes-David, Stephen M. Thielke, en Debby W.

Tsuang. 2019. “Detection of probable dementia cases

in undiagnosed patients using structured and

unstructured electronic health records”. BMC Medical

Informatics and Decision Making. https://doi.org/

10.1186/s12911-019-0846-4.

Sweeney, Elizabeth M., Russell T. Shinohara, Navid Shiee,

Farrah J. Mateen, Avni A. Chudgar, Jennifer L.

Cuzzocreo, Peter A. Calabresi, Dzung L. Pham, Daniel

S. Reich, en Ciprian M. Crainiceanu. 2013. “OASIS is

Automated Statistical Inference for Segmentation, with

applications to multiple sclerosis lesion segmentation in

MRI”. NeuroImage: Clinical. https://doi.org/10.1016/

j.nicl.2013.03.002.

Deep Learning Type Convolution Neural Network Architecture for Multiclass Classiﬁcation of Alzheimer’s Disease

215