Content uploaded by Md Milon Uddin
Author content
All content in this area was uploaded by Md Milon Uddin on Dec 02, 2022
Content may be subject to copyright.
Content uploaded by Md Ishtyaq Mahmud
Author content
All content in this area was uploaded by Md Ishtyaq Mahmud on Dec 02, 2022
Content may be subject to copyright.
978-1-6654-9299-7/22/$31.00 ©2022 IEEE
Deep Learning Based Model for Alzheimer's Disease
Detection Using Brain MRI Images
Muntasir Mamun
Department of Computer Science
University of South Dakota
South Dakota, USA
muntasir.mamun@coyotes.usd.edu
Siam Bin Shawkat
Department of Computer Science
American International University-Bangladesh
Dhaka, Bangladesh
sb.shawkat@gmail.com
Md Salim Ahammed
Department of Basic Biomedical Sciences
University of South Dakota
South Dakota, USA
mdsalim.ahammed@coyotes.usd.edu
Md Milon Uddin
Department of Electrical Engineering
The University of Texas at Tyler
Texas, USA
muddin3@patriots.uttyler.edu
Md Ishtyaq Mahmud
College of Science & Engineering
Central Michigan University
Mount Pleasant, MI 48858, USA
mahmu4m@cmich.edu
Asm Mohaimenul Islam
Department of Computer Science
University of South Dakota
South Dakota, USA
asm.islam@coyotes.usd.edu
Abstract— Alzheimer's disease (AD) is a progressive
neurodegenerative disorder that causes problems with memory,
thinking, and behavior. And with time, symptoms become severe
enough to interfere with daily activities. Although there is no cure
for the disease, a proper management strategy starting at an early
stage can help improve quality of life and potentially slow the
disease progression. In clinical research, machine learning
techniques are frequently being used in different ways to help
detect disease conditions and progressions. Magnetic resonance
imaging (MRI) is one of the best available tools that is used to
diagnose Alzheimer's disease. However, detecting very small
changes in AD brain during the early stage of the disease is
challenging. In this study, we developed deep learning-based
models for Alzheimer's detection using the 6219 MRI images
dataset. The dataset consists of images of different degrees of
demented and non-demented brains. Four deep learning models
that are utilized in this study are Convolutional Neural Network
(CNN), ResNet101, DenseNet121, and Visual Geometry Group16
(VGG16). From the analysis, we found that CNN outperformed
other models and achieved an accuracy of 97.60%, recall of 97%,
and AUC of 99.26%, with a nominal loss of 0.091.
Keywords— Alzheimer's disease, Alzheimer's diagnosis, Deep
learning model, CNN, ResNet101, VGG16.
I. INTRODUCTION
Alzheimer's disease is a neurodegenerative disorder
primarily diagnosed in elderly people that affects memory
function and cognition [1]. In 2019 alone 121,499 listed deaths
were attributed to AD. Although deaths from heart disease,
HIV, and stroke declined between 2000 and 2019, recorded
deaths from AD grew by more than 145% [2]. An estimated 16
billion hours of care were given to people with Alzheimer's or
other dementias in 2021 by more than 11 million family
members and other unpaid caregivers. According to official
statistics, Alzheimer's disease was the sixth-leading cause of
death in the United States in 2019 and the seventh-leading cause
in 2020 and 2021 [2, 3]. Alzheimer's disease accounts for 60-
80% of dementia cases [4]. The symptoms eventually become
severe enough to interfere with daily activities. Most persons
with Alzheimer's are 65 or older, and age is the most significant
risk factor. If Alzheimer's disease develops in a person younger
than 65, it is referred to as younger-onset Alzheimer's. Early-
onset Alzheimer's disease is another name for younger-onset
dementia. Alzheimer's disease in younger patients might be in
the early, medium, or late stages [4, 5]. A person with
Alzheimer's typically lives 4 to 8 years after diagnosis;
however, depending on other variables, they may survive up to
20 years if we take the proper step at the right time [4].
For detecting Alzheimer's disease, a brain MRI scan/ image
is a crucial subject for consideration [6]. Moreover, deep neural
network models are getting very popular for health diagnosis
using MRI images [7, 9].
However, with recent developments in machine learning,
critical health problems are diagnosed and getting successful
output [10, 11, 22, 23], particularly in the form of deep learning,
have made significant progress in the area of image
understanding by making it easier to recognize, categorize, and
quantify patterns in medical images.
The key to the advancements is leveraging hierarchical
feature representations learned exclusively from data instead of
handmade features primarily created based on domain-specific
expertise. Deep learning is quickly establishing itself as the
cutting-edge foundation in this approach, achieving improved
performances in various medical applications [8].
0510
2022 IEEE 13th Annual Ubiquitous Computing, Electronics & Mobile Communication Conference (UEMCON) | 978-1-6654-9299-7/22/$31.00 ©2022 IEEE | DOI: 10.1109/UEMCON54665.2022.9965730
Authorized licensed use limited to: CMU Libraries - library.cmich.edu. Downloaded on December 02,2022 at 16:35:35 UTC from IEEE Xplore. Restrictions apply.
Our work compared the performance of models using
several deep-learning models, including CNN, ResNet101,
DenseNet121, and VGG16. The following subsections made up
the remainder of the paper: Deep learning models that have
been employed to identify AD, a comprehensive overview, and
related works are provided in part II. Section III explains the
dataset, experimental concepts, and their applications with
complete methodology. The model performances are
demonstrated in Section IV. Finally, Section V provides the
conclusion and future work.
II. RELATED WORK
Sarraf et al. [12] used standard convolutional neural
networks to distinguish Alzheimer's disease from brain MRI
scans. The significance of identifying this type of medical data
lies in its capacity to construct a predictive model or system for
recognizing the symptoms of Alzheimer's disease compared to
normal patients and estimating the stages of the disease. The
authors used the "Alzheimer's disease Neuroimaging Initiative"
(ADNI) dataset having 43 images for the validation. The
authors achieved a mean accuracy of 96.85% for detecting
Alzheimer's using CNN.
Using analysis of brain MRI data, Islam et al. [13] developed
a deep convolutional neural network for diagnosing
Alzheimer's disease and comparison made with other deep pre-
trained models such as ResNet, InceptionV4, ADNet. Most of
the approaches used binary classification, and their model can
distinguish between the various phases of Alzheimer's disease
and performs better for early-stage diagnosis. According to the
resulting output, CNN outperformed other models and achieved
an accuracy of 93%, F1 score of 92%, precision of 94%, and
recall of 93%. The dataset has been collected from the OASIS
[14], and 416 Image samples were used for the implementation
using the input size of 112 x 112.
Pan et al. [15] proposed the CNN method with ensemble
techniques such as CNN+EL approach for identifying
Alzheimer's using MRI images. For the validation, authors used
ADNI dataset of 278 subjects. For the implementation authors
used PCA+SVM, 3D-SENet, CNN+EL techniques where
CNN+EL outperformed other methods and achieved an
accuracy 84 ± 5%, and AUC of 92 ± 3%.
Bae et al. [16] proposed a CNN-based model for detecting
Alzheimer's disease using MRI scan images from Alzheimer's
patients of different ethnicity, education level, and ages/gender.
For this reason, the authors used two types of datasets for
validation and accurate output. The datasets are from Seoul
National University Bundang Hospital (SNUBH) and
Alzheimer's disease Neuroimaging Initiative (ADNI). The
authors used 195 images for both datasets and achieved a mean
AUC of 0.91-0.94, a mean accuracy of 0.88-0.89, and a mean
sensitivity of 0.85-0.88. Authors reported that the mean
processing time per person takes 23-24s.
Ahmed et al. [17] proposed a simple CNN architecture for
Alzheimer's detection using MRI images focusing on both the
left and right hippocampus areas. For the validation, the authors
used two types of datasets such as Gwangju Alzheimer's and
Related Dementia (GARD) and ADNI, respectively. For
ADNI, 352 MRI and 326 MRI scans were used to implement
GARD. The authors achieved an accuracy of 90.05%.
Fong et al. [18] proposed deep learning object detection
algorithms such as Faster R-CNN, SSD, and YOLOv3 in the
area of Alzheimer's disease classification using the ADNI
dataset of 500 RAW MRI images. Without applying any MRI
pre-processing technique to the dataset, the authors obtained an
accuracy of 99.8% for YOLOv3, 98.2% for SSD, and 0.98.8%
for Faster R-CNN.
Helaly et al. [19] categorized the medical images and found
Alzheimer's disease, where two techniques were applied. In the
first technique, 2D and 3D convolution-based simple CNN
architectures were used to process structural brain scans in 2D,
and 3D from the Alzheimer's disease Neuroimaging Initiative
(ADNI) dataset having 5764 images and achieving the
accuracies of 93.61% and 95.17% respectively. The second
approach used the VGG19 model and other previously trained
medical images for classification by applying the transfer
learning principle and achieved an accuracy of 97% for multi-
class classification.
Battineni et al. [20] proposed a framework based on
supervised learning classifiers for categorizing dementia
patients as Alzheimer's or non-Alzheimer's disease based on
longitudinal brain MRI features. The authors collected 150
subjects from the OASIS dataset for validation. Six different
supervised classifiers such as- Gradient boosting, Support
vector machine, Logistic regression, Random forest,
AdaBoosting, Naive bayes were incorporated to classify AD
subjects using 10-fold cross-validation. And results represented
that the Gradient boosting algorithm outperforms other models
with 97.58% of accuracy, AUC of 98.1% and recall of 96%.
Orouskhani et al. [21] proposed a unique deep triplet
network in order to analyze brain MRI data and detect
Alzheimer's disease. Because of the small number of samples,
the suggested deep triplet network adds a conditional loss
function to increase the model's accuracy. This model's
fundamental network was inspired by VGG16 and compared
the output with six state art of models such as ResNet-CNN,
3D-DSA-CNN, 3D-CNN, GoogLeNet, SCNN, Conditional
triplet-VGG. The Conditional triplet-VGG achieved the highest
accuracy of 99.41%. For the implementation, they used 256 x
256 input images. The experiments were carried out using
open-access imaging research (OASIS) and collected 382
images of the brain MRI dataset.
0511
Authorized licensed use limited to: CMU Libraries - library.cmich.edu. Downloaded on December 02,2022 at 16:35:35 UTC from IEEE Xplore. Restrictions apply.
TABLE I Alzheimer's detection model performance analysis
Authors
(year)
Dataset
Collection
(Image
samples)
Applied Models Measures
(Proposed
model)
Sarraf et al.
(2016) [12]
ADNI (43) CNN Accuracy:
96.85%
Islam et al.
(2018) [13]
OASIS (416)
CNN
(
Proposed
),
ResNet,
InceptionV4,
ADNet
Accuracy:
93%,
F1-score:
92%,
Precision:
94%, and
recall: 93%.
Pan et al.
(2020) [15]
ADNI (278) PCA+SVM,
3D-SENet,
CNN+EL
(Proposed)
Accuracy 84
± 5%,
AUC: 92 ±
3%.
Bae et al.
(2020) [16]
SNUBH(195)
ADNI(195)
CNN Accuracy:
(0.88-0.89)%,
AUC:
(0.91-0.94)%,
Sensitivity:
(0.85-0.88)%
Ahmed et al.
(2020) [17]
GARD (326)
ADNI (352)
CNN Accuracy:
90.05%.
Fong et
al.(2020) [18]
ADNI (500) R-CNN,
SSD,
YOLOv3
(Proposed)
Accuracy:
99.8%
Helaly et al.
(2021) [19]
ADNI (5764) CNN,
VGG19
(proposed)
Accuracy:
97%.
Battineni et
al. (2021)
[20]
OASIS (150) Gradient
boosting
(Proposed),
Support vector
machine, Logistic
regression,
Random forest,
AdaBoosting,
Naive bayes
Accuracy:
97.58%,
AUC:
98.1%,
Recall:
96%.
Orouskhani et
al. (2022)
[21]
OASIS (382) ResNet-CNN,
3D-DSA-CNN,
3D-CNN,
GoogLeNet,
SCNN,
Conditional
triplet-
VGG(Proposed).
Accuracy:
99.41%.
Our work
(2022)
Kaggle (6219) CNN(Proposed),
ResNet101,
DenseNet121,
VGG16
Accuracy:
97.60%,
Recall: 97
%,
AUC:
99.26%
III. M
ETHODOLOGY
The methodology starts with the image dataset collected
from the available source. After that, we pre-process the image
dataset. The proposed CNN model and other deep learning
models, such as ResNet101, DenseNet121, and VGG16, are
then trained, tested, and validated on the MRI image dataset
using the standard hold-out-validation approach. The findings
are computed and analyzed to establish the best deep learning-
based model for detecting Alzheimer's disease. ResNet101,
DenseNet121, and VGG16 are pre-trained transfer learning
models, whereas CNN is a custom-trained model [24, 25]. For
this reason, the proposed custom CNN architecture is
represented in Figure 4, and Figure 1 depicts the overview of
the proposed strategy.
Figure 1: The overview of the study
A. Dataset Collection:
In this paper, the Alzheimer's Dataset (Brain MRI
Images) has been collected from the publicly available
Kaggle online source. According to the dataset source,
the images were hand collected from various websites,
with each and every label verified [26]. The data
consists of 6219 MRI images. The dataset has two
classes: one consists of a mixed number of Mild
Demented, Moderate Demented, and Very Mild
0512
Authorized licensed use limited to: CMU Libraries - library.cmich.edu. Downloaded on December 02,2022 at 16:35:35 UTC from IEEE Xplore. Restrictions apply.
Demented image data, and the other is non-demented
image data for detecting Alzheimer's diseases.
Figure 2: Positive MRI Alzheimer's
Figure 3: Negative MRI Alzheimer's
B. Dataset pre-processing:
Dataset pre-processing has been done by using
feature extraction for the images, such as reading the
images, resizing images, removing noises (de-noise),
image segmentation, and morphology (smoothing
edges). This processing system is crucial for analyzing
deep learning models for image classification or
detection processes.
C. Validation process:
It is crucial to select the appropriate validation
procedure for large image datasets. We used a hold-
out validation process by holding 70% data for
training 15% of the data for testing and 15% of the data
for validating. The hold-out validation method
produces the most effective results [27].
D. Proposed CNN architecture:
In the proposed CNN, the 32x32 input image was
first sent to a convolution layer with the value of 16
filters, 30x30 feature maps, and a kernel size of 3x3 to
look for the most general features. In order to reduce
the size of spatial data for the following layer by half,
the output of the convolutional layer was then passed
on to a max pooling layer with feature maps of 15x15.
This output was then sent to a second convolution
layer with the value of 32 filters, feature maps of
13x13, and kernel size set to 3x3 for further
processing. To reduce the quantity of spatial data for
the following layer by half, the output of this layer was
then passed on to a max pooling layer with 6x6 feature
maps. Another set of convolution and pooling layers
comes with this. In this instance, the convolution layer
was made up of 64 filters with 2x2 feature maps and a
kernel size of 3x3, and the pooling layer was made up
of 1x1 feature maps. The final output last
convolutional layer was then flattened and passed to
130 dimensional fully connected dense layer that was
created. It is then routed to the final output layer,
which has a softmax activation function. All layers use
a ReLU activation function with a dropout of 0.5,
except for the final layer, which uses a softmax
activation with no dropout. Figure 4 depicts the above-
mentioned layout of the proposed CNN architecture.
The model was trained with a learning rate of 0.01, 50
epochs, and 13 batch sizes. The adamax optimizer was
used for compiling the model. A categorical
crossentropy-based loss function and other metrics
such as accuracy, recall, and AUC has been achieved
using the help of the Keras python library.
Figure 4: CNN architecture
0513
Authorized licensed use limited to: CMU Libraries - library.cmich.edu. Downloaded on December 02,2022 at 16:35:35 UTC from IEEE Xplore. Restrictions apply.
IV. R
ESULTS
A
ND
D
ISCUSSION
The findings of different types of advanced deep learning
models - i.e. CNN, ResNet101, DenseNet121, VGG16
classification algorithms on Alzheimer's image dataset has been
computed in TABLE II, and comparisons have been provided
in Figures 5. In TABLE II, we presented the accuracy, area
under the curve (AUC), recall, and loss function results for the
performance observation of the models.
Table II. Results of different measures for different deep learning models for
detecting Alzheimer's diseases.
Model Accuracy AUC Recall Loss
1. CNN 97.60% 99.26% 97% 0.091
2. ResNet101 73.85% 83.00% 73.80% 0.556
3. DenseNet121 72.00% 77.63% 71.60% 0.640
4. VGG16 70.20% 77.10% 70.00% 0.583
In Figure 5, we can observe that CNN achieved the highest
accuracy of 97.60%, whereas VGG16 achieved the least
accuracy of 70.20%. Apart from that, ResNet101 and
DenseNet121 attained an accuracy of 73.85% and 72%,
respectively.
Figure 5: Deep Learning model's performance analysis in
terms of Accuracy and AUC
However, accuracy alone cannot clearly show a sufficient
measurement system for evaluating a model's performance.
Besides, the AUC value and loss function become a vital matrix
for determining the model's performance and assessing a
model's ability to differentiate between classes. The AUC
measures how well the model differentiates between positive
and negative classes. The higher the AUC value, the better. The
value range is 0 to 1, with 0 representing a completely
inaccurate test and 1 representing a completely accurate test. In
general, an AUC of 0.5 indicates no discrimination (i.e., the
ability to classify patients with and without Alzheimer's
diseases or condition based on the test), 0.7 to 0.8 is considered
acceptable, 0.8 to 0.9 is considered great, and larger than 0.9 is
considered outstanding performance [28]. According to Figure
5, CNN not only achieved the highest accuracy but also
achieved an outstanding AUC score, which is 99.26%. Besides
ResNet101, DenseNet121, and VGG16 achieved an AUC score
of 83%, 77.63% and 77.10% respectively. In addition, the loss
function is another considerable metric for evaluating the
model's performance. The loss function tells us about the
deviation between validation and training values. If the
validation value is too much away from the training value, then
our loss function's values tend to be high. So lowering the loss
value helps the model for better performance. A categorical
crossentropy loss function was analyzed along with the adam
optimizer. From Table II, we can see that CNN achieved the
lowest loss value of 0.091 whereas, DenseNet121 achieved the
highest loss of 0.640 according to our analysis. From the overall
analysis, the custom CNN model performed excellently in
detecting Alzheimer's disease using the brain MRI image
dataset and outperformed other deep learning models. We
provided the validation accuracy, validation AUC and loss
function curve in every epoch for CNN in Figures 6, 7, and 8,
respectively.
Figure 6: Accuracy curve for CNN
97.60%
73.85% 72% 70.20%
99.26%
83% 77.63% 77.10%
0
20
40
60
80
100
120
Accuracy AUC
0514
Authorized licensed use limited to: CMU Libraries - library.cmich.edu. Downloaded on December 02,2022 at 16:35:35 UTC from IEEE Xplore. Restrictions apply.
Figure 7: AUC curve for CNN
Figure 8: Loss function curve for CNN
V. C
ONCLUSION AND
F
UTURE WORK
A common early sign of Alzheimer's is significant memory
loss, disorientation, and other significant changes in our mental
functioning may be symptoms of failing brain cells. With time,
these symptoms worsen, and the quality of life degrades.
Alzheimer's has no cure, but proper management in a timely
fashion can potentially improve quality of life and delay disease
progression. Deep learning-based models are used for image
classification in health care, and Brain MRI images can be one
of the
most effective datasets for the detection process. This
paper analyzed four types of deep architectures, such as CNN,
ResNet101, DenseNet121, and VGG16, to detect AD using
brain MRI images. According to the overall analysis, CNN
outperformed other models and is
considered the proposed
model for Alzheimer's detection. CNN achieved the highest
accuracy of 97.60%, recall of 97%, and AUC of 99.26%, along
with a nominal loss of 0.091. Our developed models, especially
CNN,
achieved promising results. This proposed model might
be beneficial for early diagnosis and therapy of Alzheimer's and
contribute to biomedical research. In the future, we may use
newly developed deep learning models and pre-trained deep
architectures for more accurate results for Alzheimer's
detection. In the same way, we will also work on cancer
detection and some other diseases using deep learning
algorithms for humankind.
R
EFERENCES
[1] Ahammed S, Afrin R, Uddin N, Al-Amin Y, Hasan K, Haque U, Islam
KMM, Alam AHMK, Tanaka T, Sadik G. Acetylcholinesterase Inhibitory
and Antioxidant Activity of the Compounds Isolated from Vanda
roxburghii. Adv Pharmacol Pharm Sci. 2021 Mar 27;2021:5569054. doi:
10.1155/2021/5569054. PMID: 33855299; PMCID: PMC8019628.
[2] (2022), 2022 Alzheimer's disease facts and figures. Alzheimer's Dement.,
18: 700-789. https://doi.org/10.1002/alz.12638
[3] (2020), 2020 Alzheimer's disease facts and figures. Alzheimer's Dement.,
16: 391-460. https://doi.org/10.1002/alz.12068
[4] Alzheimer's Disease and Dementia. 2022. What is Alzheimer's?. [online]
Available at: <https://www.alz.org/alzheimers-dementia/what-is-
alzheimers> [Accessed 5 September 2022].
[5] 2022. [online] Available at: <https://www.nia.nih.gov/health/alzheimers-
disease-fact-sheet> [Accessed 5 September 2022].
[6] Teipel, S., Grothe, M., Lista, S., Toschi, N., Garaci, F. and Hampel, H.,
2013. Relevance of Magnetic Resonance Imaging for Early Detection and
Diagnosis of Alzheimer Disease. Medical Clinics of North America,
97(3), pp.399-424.
[7] Saleem, T.J.; Zahra, S.R.; Wu, F.; Alwakeel, A.; Alwakeel, M.; Jeribi, F.;
Hijji, M. Deep Learning-Based Diagnosis of Alzheimer's Disease. J. Pers.
Med. 2022, 12, 815. https://doi.org/10.3390/jpm12050815
[8] Shen D, Wu G, Suk HI. Deep Learning in Medical Image Analysis. Annu
Rev Biomed Eng. 2017 Jun 21;19:221-248. doi: 10.1146/annurev-bioeng-
071516-044442. Epub 2017 Mar 9. PMID: 28301734; PMCID:
PMC5479722.
[9] Ebrahimi, A., Luo, S. and Chiong, R., 2021. Deep sequence modelling for
Alzheimer's disease detection using MRI. Computers in Biology and
Medicine, 134, p.104537.
[10] M. Mamun, A. Farjana, M. Al Mamun and M. S. Ahammed, "Lung cancer
prediction model using ensemble learning techniques and a systematic
review analysis," 2022 IEEE World AI IoT Congress (AIIoT), 2022, pp.
187-193, doi: 10.1109/AIIoT54504.2022.9817326.
[11] M. Mamun, A. Farjana, M. A. Mamun, M. S. Ahammed and M. M.
Rahman, "Heart failure survival prediction using machine learning
algorithm: am I safe from heart failure?," 2022 IEEE World AI IoT
Congress (AIIoT), 2022, pp. 194-200, doi:
10.1109/AIIoT54504.2022.9817303.
[12] S. Sarraf and G. Tofighi, "Deep learning-based pipeline to recognize
Alzheimer's disease using fMRI data," 2016 Future Technologies
Conference (FTC), 2016, pp. 816-820, doi: 10.1109/FTC.2016.7821697.
[13] Islam, J. and Zhang, Y., 2018. Brain MRI analysis for Alzheimer's disease
diagnosis using an ensemble system of deep convolutional neural
networks. Brain Informatics, 5(2).
0515
Authorized licensed use limited to: CMU Libraries - library.cmich.edu. Downloaded on December 02,2022 at 16:35:35 UTC from IEEE Xplore. Restrictions apply.
[14] Marcus DS, Wang TH, Parker J, Csernansky JG, Morris JC, Buckner RL.
Open Access Series of Imaging Studies (OASIS): cross-sectional MRI
data in young, middle aged, nondemented, and demented older adults. J
Cogn Neurosci. 2007 Sep;19(9):1498-507. doi:
10.1162/jocn.2007.19.9.1498. PMID: 17714011.
[15] Pan D, Zeng A, Jia L, Huang Y, Frizzell T and Song X (2020) Early
Detection of Alzheimer's Disease Using Magnetic Resonance Imaging: A
Novel Approach Combining Convolutional Neural Networks and
Ensemble Learning. Front. Neurosci. 14:259.doi:
10.3389/fnins.2020.00259
[16] Bae, J., Lee, S., Jung, W., Park, S., Kim, W., Oh, H., Han, J., Kim, G.,
Kim, J., Kim, J. and Kim, K., 2020. Identification of Alzheimer's disease
using a convolutional neural network model based on T1-weighted
magnetic resonance imaging. Scientific Reports, 10(1).
[17] S. Ahmed et al., "Ensembles of Patch-Based Classifiers for Diagnosis of
Alzheimer Diseases," in IEEE Access, vol. 7, pp. 73373-73383, 2019, doi:
10.1109/ACCESS.2019.2920011.
[18] J. X. Fong, M. I. Shapiai, Y. Y. Tiew, U. Batool and H. Fauzi, "Bypassing
MRI Pre-processing in Alzheimer's Disease Diagnosis using Deep
Learning Detection Network," 2020 16th IEEE International Colloquium
on Signal Processing & Its Applications (CSPA), 2020, pp. 219-224, doi:
10.1109/CSPA48992.2020.9068680.
[19] Helaly, H.A., Badawy, M. & Haikal, A.Y. Deep Learning Approach for
Early Detection of Alzheimer's Disease. Cogn Comput (2021).
https://doi.org/10.1007/s12559-021-09946-2
[20] Battineni, G.; Hossain, M.A.; Chintalapudi, N.; Traini, E.; Dhulipalla,
V.R.; Ramasamy, M.; Amenta, F. Improved Alzheimer's Disease
Detection by MRI Using Multimodal Machine Learning Algorithms.
Diagnostics 2021, 11,
2103.https://doi.org/10.3390/diagnostics11112103.
[21] Orouskhani, M., Zhu, C., Rostamian, S., Shomal Zadeh, F., Shafiei, M.
and Orouskhani, Y., 2022. Alzheimer's disease detection from structural
MRI using conditional deep triplet network. Neuroscience Informatics,
2(4), p.100066.
[22] M. Mamun, M. M. Uddin, V. K. Tiwari, A. M. Islam, A. U. Ferdous,
"MLHeartDis: Can Machine Learning Techniques Enable to Predict
Heart Diseases?", 2022 IEEE 13th Annual Ubiquitous Computing,
Electronics & Mobile Communication Conference (UEMCON), 2022,
(Preprint)
[23] M. Mamun, M. I. Mahmud, M. I. Hossain, A. M. Islam, M. S. Ahammed,
M. M. Uddin, "Vocal Feature Guided Detection of Parkinson's Disease
Using Machine Learning Algorithms", 2022 IEEE 13th Annual
Ubiquitous Computing, Electronics & Mobile Communication
Conference (UEMCON), 2022, (Preprint)
[24] K. S. a. A. Zisserman, "Very deep convolutional networks for large-scale
image recognition,," in International Conference on Learning
Representations (ICLR), San Diego, 2015.
[25] Russakovsky, O., Deng, J., Su, H., et al. "ImageNet Large Scale Visual
Recognition Challenge." International Journal of Computer Vision
(IJCV). Vol 115, Issue 3, 2015, pp. 211–252
[26] Dubey, S., 2022. Alzheimer's Dataset ( 4 class of Images). [online]
Kaggle.com. Available at:
<https://www.kaggle.com/datasets/tourist55/alzheimers-dataset-4-class-
of-images> [Accessed 1 September 2022].
[27] Dwork, C., Feldman, V., Hardt, M., Pitassi, T., Reingold, O., & Roth, A.
(2015). The reusable holdout: Preserving validity in adaptive data
analysis. Science, 349(6248), 636-638. doi: 10.1126/science.aaa9375
[28] Mandrekar, J. (2010). Receiver Operating Characteristic Curve in
Diagnostic Test Assessment. Journal Of Thoracic Oncology,5(9), 1315-
1316.doi: 10.1097/jto.0b013e3181ec173d
0516
Authorized licensed use limited to: CMU Libraries - library.cmich.edu. Downloaded on December 02,2022 at 16:35:35 UTC from IEEE Xplore. Restrictions apply.
