A modified residual network for detection and classification of Alzheimer’s disease

International Journal of Electrical and Computer Engineering (IJECE)
Vol. 12, No. 4, August 2022, pp. 4400~4407
ISSN: 2088-8708, DOI: 10.11591/ijece.v12i4.pp4400-4407  4400
Journal homepage: http://ijece.iaescore.com
A modified residual network for detection and classification of
Alzheimer’s disease
Faten Salim Hanoon, Abbas Hanon Hassin Alasadi
Computer Information Systems Department Computer Science and Information Technology College, University of Basrah, Basrah, Iraq
Article Info ABSTRACT
Article history:
Received Jun 23, 2021
Revised Feb 12, 2022
Accepted Mar 1, 2022
Alzheimer's disease (AD) is a brain disease that significantly declines a
person's ability to remember and behave normally. By applying several
approaches to distinguish between various stages of AD, neuroimaging data
has been used to extract different patterns associated with various phases of
AD. However, because the brain patterns of older adults and those in
different phases are similar, researchers have had difficulty classifying them.
In this paper, the 50-layer residual neural network (ResNet) is modified by
adding extra convolution layers to make the extracted features more diverse.
Besides, the activation function (ReLU) was replaced with (Leaky ReLU)
because ReLU takes the negative parts of its input, drops them to zero, and
retains the positive parts. These negative inputs may contain useful feature
information that could aid in the development of high-level discriminative
features. Thus, Leaky ReLU was used instead of ReLU to prevent any
potential loss of input information. In order to train the network from scratch
without encountering the issue of overfitting, we added a dropout layer
before the fully connected layer. The proposed method successfully
classified the four stages of AD with an accuracy of 97.49 % and 98 % for
precision, recall, and f1-score.
Keywords:
Alzheimer's disease
CNN
Deep learning
Medical imaging
Transfer learning
This is an open access article under the CC BY-SA license.
Corresponding Author:
Abbas Hanon Hassin Alasadi
Faculty of Computer Science and Information Technology, University of Basrah
Basrah, Iraq
Email: abbas.hassin@uobasrah.edu.iq
1. INTRODUCTION
Alzheimer's disease (AD) is a neurological disease. Its symptoms are a decline in memory, cognitive
skill impairment, lack of reasoning and judgment, and difficulties with expression and comprehension [1].
The main reason for Alzheimer's disease is that the brains of patients have higher levels of amyloid [2], [3]
and tau protein than the normal brains of older people [4], [5]. Accordingly, the increasing growth of amyloid
plaques and tau neurofibrillary tangles leads to a loss of nerve cells' ability to communicate with one another
and, eventually, cell death. The first brain area to be negatively impacted by a disease is the hippocampus [6].
Due to the hippocampus's critical role in learning and memory, forgetting people and events is the first sign
of AD. The number of people who have Alzheimer's disease Precipitate will quickly grow as life expectancy
grows. According to the world Alzheimer's report [7] statistics, an estimated 50 million people had Alzheimer's
disease in 2015. By 2050, this number is expected to increase to 131.5 million people worldwide.
Presently, no treatment or drug can delay or stop the growth of Alzheimer's disease, necessitating
and requiring effective and precise methods for early detection and preventing the disease from progressing
to late stages. The practice of assigning items to a specific set based on their characteristics is called
classification [8]. One of the specific goals of artificial intelligence research is to classify diseases [9]. A
variety of neuroimaging methods are be of assistance in the classification of AD, for instance, magnetic

Int J Elec & Comp Eng ISSN: 2088-8708 
A modified residual network for detection and classification of Alzheimer’s disease (Faten Salim Hanoon)
4401
resonance imaging (MRI), computerized tomography (CT), functional MRI (fMRI), positron emission
tomography (PET), magnetoencephalography (MEG), electroencephalography (EEG). MRI technologies are
the most commonly used because images obtained by MRI show the affected cells darker than the healthy
regions [10]. Figure 1 depicts several brain MRI images illustrating various stages of Alzheimer's disease.
Dechter established the concept of deep learning (DL) in 1986 [11]. Its techniques, especially convolutional
neural networks (CNN), have gained traction in a variety of fields, including image processing and analysis,
and have achieved good performance in many computer vision tasks and medical imaging applications such
as cerebral microbleeds (CMBs) detection [12].
This paper presents a modified version of the 50-layer residual neural network (ResNet) architecture
that can solve the classification problem of four stages of Alzheimer's disease. Alzheimer's disease
progressively deteriorates brain tissue in a predictable pattern. It reduces the size of the hippocampus and
cerebral cortex of the brain while increasing the size of the ventricles [13]. Some outstanding research has
been conducted on automated Alzheimer's disease diagnosis. The method of Jongkreangkrai et al. [14]
consisted of two essential phases: the feature extraction phase, where they used MRI brain images to extract
feature sets containing volumes of the hippocampus and amygdala as well as the thickness of the entorhinal
cortex. After that, they moved on to the classification stage. They used the support vector machine algorithm
to differentiate between Alzheimer's patients and healthy subjects based on their extracted features. Since the
amount of medical data available is inadequate in comparison to other fields, as well as the fact that deep
neural network training necessitates a large number of computing resources. Transfer learning is a promising
alternative methodology developed by researchers. Jain et al. [15] used a pre-trained VGG16 network on the
ImageNet dataset as a feature extractor for classifying brain MRI images into three categories: CN, MCI,
MD, and AD as shown in Figure 1(a) to 1(d). Odusami et al. [16] introduced a new brain slice classification
approach based on the ResNet18 algorithm. Puente-Castro et al. [10] used the first 47 layers of ResNet to
extract features from sagittal MRI images. They then added the patient's age and gender to the feature vectors
extracted by ResNet.
(a) (b) (c) (d)
Figure 1. Brain MRI images for various stages of Alzheimer's disease: (a) cognitively normal (CN),
(b) mild demented (MCI), (c) moderate demented (MD), and (d) very mild demented (AD)
The support vector machine (SVM) technique is then used as a classifier on these extracted feature
vectors to determine whether the patient is in any stage of Alzheimer's disease or CN. Rabeh et al. [17]
proposed an application for the early detection of Alzheimer's disease. The application framework has two
steps: segmentation using region of interest (ROI) to isolate three critical regions: hippocampus, corpus
callosum, and cortex. Following that, there is a classification step for components by using SVM and
applying a decision tree to make the final decision. Ji et al. [18] developed a method for early diagnosis of
Alzheimer's disease using MRI images of human brains and ResNet50, NASNet and MobileNet as base
classifiers, which were trained in an end-to-end process
2. THE PROPOSED METHOD
Accuracy in medical diagnosis is more important than anything else, even more, important than
speed of diagnosis. After all, the wrong diagnosis of an ordinary person as a patient causes severe
consequences and psychological pressure, as well as diagnosing a patient, as usual, leads to the development
of the disease because the wrong diagnosis, in this case, will delay treatment. Therefore, building an
automated medical diagnostic system must be at a high level of accuracy.
2.1. Convolutional neural networks (CNN)
Convolutional neural networks are used in various applications, including image classification,
segmentation, and pattern recognition [19], [20]. Due to its autonomous nature has developed into a critical
tool for machine vision and artificial intelligence. CNN is a particular neural network that directly applies

 ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 12, No. 4, August 2022: 4400-4407
4402
image processing to the pixels without prior treatments [21]. The CNN architecture consists of three primary
layers: the convolution layer, the pooling layer, and the fully connected layer. The convolution layer is the
building block of the CNN algorithm. It is in charge of extracting the essential and beneficial features from
the input images using a set of trainable filters, forming a feature map [22]. Between successive
convolutions is a pooling layer used to reduce the feature mapping dimensions in computational space [23],
thereby lowering the computational cost of the subsequent convolution layer, which contributes to the
acceleration of training and the enhancement of generalization ability [24]. Typically, fully connected layers
are inserted at the end of the CNN structure, utilized for recognition and classification. Every node in the
fully connected layers is linked with trainable weights in adjacent fully connected layers. The fully
connected layer produces a final output equal to the number of classes. CNN is a stacked version of all the
layers that make up the CNN architecture. Each CNN, with a few exceptions, uses the same architecture.
Fully training a new CNN from scratch is not without its challenges. Firstly, CNN needs large
amounts of labeled data for the training process that may be difficult to obtain, especially in medical
imaging. In addition, to train a CNN, need to put in lots of computing and memory resources. Otherwise, the
training process would take a long time without these resources. Tuning hyperparameters is time-consuming
and complicated, and it can result in overfitting or underfitting, which leads to poor model performance.
Researchers have demonstrated a promising alternative method known as transfer learning to overcome
these obstacles. Transfer learning means improving the learning of a new task through transferring
knowledge from a previously learned task [25]. The fundamental goal of this research is to compare the
performance of pre-trained ResNet50 against the modified residual network in detecting and automatically
classifying Alzheimer's disease using MRI scans.
2.2. Proposed framework
This section presents the general architecture of the proposed framework. The proposed framework
consists of the following stages: data collection, data preparation, model selection, training stage to build the
model that helps in diagnosis, validation, and evaluation. Each stage is independent of the other and is
responsible for implementing a specific function. At the same time, these stages can communicate with each
other since the result of one stage will be the input to the different stages. Figure 2 describes the proposed
framework.
Figure 2. Flowchart of the proposed framework for early diagnosis of Alzheimer’s disease

4403
2.2.1. Data preparation
Data collection is the first step in the machine learning pipeline for training the selected model. The
accuracy of ML systems' predictions is only good when the data used to train them is good. So, the first
stage in the framework of the proposed work in this work is to collect data and obtain it from data sources
related to this work in order to solve the research problem and evaluate the results. In this study, the
Alzheimer's brain MRI dataset was obtained from the open access of the Kaggle website. The dataset
contains 6400 images with a size of 176×208 pixels. It has four classes (NonDemented, MildDemented,
ModerateDemented, VeryMildDemented) with a non-uniform distribution of the images per class.
The second step is changing size of image to (125×125) in order to decrease the time of training the
neural network by lowering the number of pixels in an image because the more pixels in an image, the more
input nodes there are, which raises the model's complexity. After that and because of the non-uniform
distribution of images in each class, we generated new training examples using one of the data augmentation
techniques only on the training set to improve deep neural network generalization capabilities and prevent
overfitting. Horizontal flipping is the augmentation technique used in this step. This technique works by
shifting all of the pixels in an image in the horizontal direction, or other words, by reversing the entire rows
and columns of image pixels horizontally.
The data set is separated into two independent sets; the training set which has (5,121) images and
the testing set which has (1,279) images. The training set was then partitioned into into a 20% validation
dataset and an 80% new training dataset. The proposed model is trained using the “new training set”. In
contrast, the validation set periodically evaluates the model's performance during the training phase to avoid
overfitting problems. The testing set is later used to evaluate how well the model generalizes to unseen data.
Shuffling is the last step in data preparation. It regulates the weights, which means getting lower weights
closer to zero. The most critical aspect is preventing the model from learning the training order. This step
eventually helps the training converge quickly so that the network can provide better generalizations.
During the validation phase and testing phase, there is no shuffling process for the model's parameters.
During the validation and testing phases, we calculate accuracy and loss. Their calculation method is not
sensitive to the order of samples, so shuffling does not affect the testing and validation data.
2.2.2. Network architecture
The residual neural network, known as ResNet, is a deep neural network that uses shortcuts, called
"skip connections", to jump over some layers [26], [27]. ResNet has demonstrated outstanding performance
in computer vision, so that it will be used in current research. He et al. [28] invented the ResNet in 2016,
and it earned first place in the ILSVRC 2015 classification competition with a 3.57 error rate.
ResNet's hypothesis is that deeper networks are more difficult to optimize, as the deeper model
should be capable of performing as well as the shallower model by copying the shallower model's learned
parameters and setting additional layers for identity mapping. To aid in the optimization of deeper models,
residual blocks are designed to fit a residual mapping F (x) rather than the desired underlying mapping H (x)
to assist in the optimization of deeper models, and entire ResNet architectures are built by stacking residual
blocks. Figure 3 illustrates the concept of a residual block. If we assume that the input is x, the convolution
layer's output is F (x), which is added to x as the mapping input, and the resulting output H (x) = F (x) + x is
passed to the next layer. This is significantly easier than matching an identity map through a collection of
nonlinear layers, and it does not require the network to include additional parameters and calculations.
Simultaneously, it can significantly increase the training speed and effectiveness of the model as the number
of layers increases. This residual block structure can effectively solve the problem of gradient vanishing in
deep networks [28]. There are two types of residual blocks in ResNet. While the first type is suitable for
training shallow networks, the second type (bottleneck) is recommended for more than 50 layers.
Additionally, the two types share a similar level of time complexity.
In this paper, an improved residual neural network based on ResNet50 (named ResNetF) has
proposed. The number of convolution layers increased to 58 layers. More generally, as the number of
network layers increased, the features that are extracted from the different layers become more diverse and
richer. Additionally, the more deeply embedded the network is, the more abstract the features that are
extracted. As a result of improving the network's feature extraction abilities, its effectiveness in AD
diagnosis has improved. To ensure that overfitting is effectively avoided. We added a dropout layer before
the fully connected layer in this architecture and set the dropout ratio to 50%. In ResNet50 ReLU is
commonly used as an activation function. Basically, in CNN, ReLU takes the negative parts of its input and
drops them to zero, and retains the positive parts. However, these negative inputs may contain useful feature
information that could be used to aid in the development of discriminative high-level features [29].

 ISSN: 2088-8708
4404
Figure 3. Structure of bottleneck block (for ResNet 50, 101,152)
If a neuron's output is 0, its gradient will never update the neuron's weight, resulting in the neuron
never being activated. When the network contains a high number of inactive neurons, the convergence of
the model becomes extremely difficult. Accordingly, this is referred to as the “dying ReLU” problem. This
may prevent the network from learning and result in underperformance. To address this problem, Leaky
ReLU is used instead of ReLU to prevent any potential loss of input information. Leaky Rectified Linear
Activation (LReLU) has added an alpha parameter to the semi-axis of ReLU, resulting in a small gradient
but not zero. Nodes that were previously inactive with ReLU will now have their weight-adjusted. Figure 4
shows our enhanced network. ReLU is given by (1), (2):
𝑓(𝑥𝑖) = 𝑚𝑎𝑥( 0, 𝑥𝑖) = 𝑓(𝑥) = {
𝑥𝑖 𝑥𝑖 > 0
0 𝑥𝑖 ≤ 0
(1)
𝑓′
(𝑥) = {
1 𝑥𝑖 > 0
0 𝑥𝑖 ≤ 0
(2)
Leaky ReLU is defined as (3), (4):
𝑓(𝑥𝑖) = {
𝑥𝑖 𝑥𝑖 > 0
𝑎𝑖𝑥𝑖 𝑥𝑖 ≤ 0
(3)
𝑓′
(𝑥) = {
1 𝑥𝑖 > 0
𝑎𝑖𝑥𝑖 𝑥𝑖 ≤ 0
(4)
where ai is a predefined parameter that falls in the range of (0,1). It is usually 0.01, xi is the activation
function's input.
Figure 4. Modified residual neural network (ResNetF) structure

4405
2.2.3 Training stage
In this stage, we used training set and the backpropagation algorithm to train our proposed network
(ResNetF). We also used Adam as the optimizer with a learning rate of 1e-5, cross-entropy as the loss
function. After each epoch during the training phase we assessed the model's performance by computing the
validation accuracy and validation loss using our validation set.
2.2.4. Evaluation Stage
In this stage, popular criteria are used to assess the output of machine learning models to measure
the classification model's ability to classify the testing dataset accurately. Accuracy, precision, recall, and
F1-score are the common measurement tools, as mentioned in (5) to (8), respectively. True positive, false
positive, and false negative predictions for a given class label are represented by the variables TP, FP, and
FN, respectively [30], [31].
𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 =
𝑇𝑃+𝑇𝑁
𝑇𝑁+𝑇𝑃+𝐹𝑃+𝐹𝑁
(5)
Precision =
𝑇𝑃
𝑇𝑃+𝐹𝑃
(6)
Recall =
𝑇𝑃
𝑇𝑃+𝐹𝑁
(7)
𝐹1 − 𝑠𝑐𝑜𝑟𝑒 = 2 × 𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 × 𝑅𝑒𝑐𝑎𝑙𝑙
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 + 𝑅𝑒𝑐𝑎𝑙𝑙
(8)
3. RESULTS AND DISCUSSION
This section explains the results of the research and, at the same time, gives a comprehensive
discussion. The proposed network was trained for 13 hours, 54 minutes, 53 seconds. The Pytorch framework
and Python 3.7.10 were used to run experiments on an Nvidia Tesla P100 GPU with 25 GB of memory.
To evaluate the classification model using testing data, various criteria are used, including precision,
recall, the f1-measure, and accuracy to assess all stages of Alzheimer's disease. According to the results of
our experiment, the training accuracy was 99%, and the validation accuracy was 97%. The results of the
evaluation are summarized in Table 1.
The proposed model has been compared to a previous study [32] that addressed a similar issue and
used the same data samples. We noticed that the model achieves encouraging results and surpasses the
previous state-of-the-art in all criteria. The success can be attributed to expanding the number of
convolutional layers, adding a dropout layer, and substituting an activation function (Relu) for an activated
function (leaky Rule) in our modified network.
Table 1. Performance of the ResNetF model
Class Model Accuracy Precision Recall F1-score
Non Demented Yildirim and Cinar [32] 90 90 96.42 93.09
ResNetF 96.56 99 97 98
Mild Demented Yildirim and Cinar [32] 96.6 96.6 90.62 93.51
ResNetF 96.64 99 97 98
Moderate Demented Yildirim and Cinar [32] 70 70 70 70
ResNetF 100 100 100 100
Very Mild Demented Yildirim and Cinar [32] 90 90 90 90
ResNetF 99 95 99 97
4. CONCLUSION
An accurate diagnosis of Alzheimer's disease allows the patient to receive the most appropriate
treatment. This challenging task focuses on many researchers, who have built up many computer-aided
diagnosis (CAD) systems to diagnose AD. This paper presented an enhanced residual neural network to
classify four stages of Alzheimer's disease. By increasing the number of convolution layers, the network can
effectively capture as many AD biomarkers as possible. In addition, substituting Relu for the activation
function (Leaky Relu) can solve losing valuable features that could assist in the construction of high-level
discriminative features. To avoid overfitting, we added a dropout layer before the fully connected layers to
train all our architecture layers from scratch. The findings of the experiments demonstrate that the enhanced
residual neural network is suitable for Alzheimer's disease diagnosis.

 ISSN: 2088-8708
4406
REFERENCES
[1] N. J. Herzog and G. D. Magoulas, “Brain Asymmetry detection and machine learning classification for diagnosis of early
dementia,” Sensors, vol. 21, no. 3, Jan. 2021, doi: 10.3390/s21030778.
[2] D. J. Selkoe and J. Hardy, “The amyloid hypothesis of Alzheimer’s disease at 25 years,” EMBO Molecular Medicine, vol. 8, no.
6, pp. 595–608, Jun. 2016, doi: 10.15252/emmm.201606210.
[3] W. Qiang, W.-M. Yau, J.-X. Lu, J. Collinge, and R. Tycko, “Structural variation in amyloid-β fibrils from Alzheimer’s disease
clinical subtypes,” Nature, vol. 541, no. 7636, pp. 217–221, Jan. 2017, doi: 10.1038/nature20814.
[4] B. Eftekharzadeh et al., “Tau protein disrupts nucleocytoplasmic transport in Alzheimer’s disease,” Neuron, vol. 99, no. 5, pp.
925–940, Sep. 2018, doi: 10.1016/j.neuron.2018.07.039.
[5] S. S. Khan and G. S. Bloom, “Tau: the center of a signaling nexus in Alzheimer’s disease,” Frontiers in Neuroscience, vol. 10,
Feb. 2016, doi: 10.3389/fnins.2016.00031.
[6] A. D. Bachstetter et al., “Disease-related microglia heterogeneity in the hippocampus of Alzheimer’s disease, dementia with
Lewy bodies, and hippocampal sclerosis of aging,” Acta Neuropathologica Communications, vol. 3, no. 1, Dec. 2015, doi:
10.1186/s40478-015-0209-z.
[7] M. Prince, A. Wimo, M. Guerchet, A. Gemma-Claire, Y.-T. Wu, and M. Prina, “World Alzheimer report 2015: the global impact
of dementia - an analysis of prevalence, incidence, cost and trends,” 2015. doi: 10.1111/j.0963-7214.2004.00293.x.
[8] H. A. Lafta, Z. F. Hasan, and N. K. Ayoob, “Classification of medical datasets using back propagation neural network powered by
genetic-based features elector,” International Journal of Electrical and Computer Engineering (IJECE), vol. 9, no. 2, pp. 1379–
1384, Apr. 2019, doi: 10.11591/ijece.v9i2.pp1379-1384.
[9] A. A. Hussein, “Improve the performance of k-means by using genetic algorithm for classification heart attack,” International
Journal of Electrical and Computer Engineering (IJECE), vol. 8, no. 2, pp. 1256–1261, Apr. 2018, doi:
10.11591/ijece.v8i2.pp1256-1261.
[10] A. Puente-Castro, E. Fernandez-Blanco, A. Pazos, and C. R. Munteanu, “Automatic assessment of Alzheimer’s disease diagnosis
based on deep learning techniques,” Computers in Biology and Medicine, vol. 120, May 2020, doi:
10.1016/j.compbiomed.2020.103764.
[11] H. Abdulkarim and M. Z. Al-Faiz, “Online multiclass EEG feature extraction and recognition using modified convolutional
neural network method,” International Journal of Electrical and Computer Engineering (IJECE), vol. 11, no. 5, pp. 4016–4026,
Oct. 2021, doi: 10.11591/ijece.v11i5.pp4016-4026.
[12] Q. Dou et al., “Automatic detection of cerebral microbleeds from MR images via 3D convolutional neural networks,” IEEE
Transactions on Medical Imaging, vol. 35, no. 5, pp. 1182–1195, May 2016, doi: 10.1109/TMI.2016.2528129.
[13] S. Sarraf, D. D. DeSouza, J. Anderson, and G. Tofighi, “DeepAD: Alzheimer’s disease classification via deep convolutional
neural networks using MRI and fMRI,” bioRxiv, pp. 1–32, 2016, doi: 10.1101/070441.
[14] C. Jongkreangkrai, Y. Vichianin, C. Tocharoenchai, and H. Arimura, “Computer-aided classification of Alzheimer’s disease
based on support vector machine with combination of cerebral image features in MRI,” Journal of Physics: Conference Series,
vol. 694, Mar. 2016, doi: 10.1088/1742-6596/694/1/012036.
[15] R. Jain, N. Jain, A. Aggarwal, and D. J. Hemanth, “Convolutional neural network based Alzheimer’s disease classification from
magnetic resonance brain images,” Cognitive Systems Research, vol. 57, pp. 147–159, Oct. 2019, doi:
10.1016/j.cogsys.2018.12.015.
[16] M. Odusami, R. Maskeliūnas, R. Damaševičius, and T. Krilavičius, “Analysis of features of Alzheimer’s disease: detection of
early stage from functional brain changes in magnetic resonance images using a finetuned ResNet18 network,” Diagnostics, vol.
11, no. 6, Jun. 2021, doi: 10.3390/diagnostics11061071.
[17] A. B. Rabeh, F. Benzarti, and H. Amiri, “Diagnosis of Alzheimer diseases in early step using SVM (support vector machine),” in
2016 13th International Conference on Computer Graphics, Imaging and Visualization (CGiV), Mar. 2016, pp. 364–367, doi:
10.1109/CGiV.2016.76.
[18] H. Ji, Z. Liu, W. Q. Yan, and R. Klette, “Early diagnosis of Alzheimer’s disease using deep learning,” in Proceedings of the 2nd
International Conference on Control and Computer Vision - ICCCV 2019, 2019, pp. 87–91, doi: 10.1145/3341016.3341024.
[19] M. Al-Smadi, M. Hammad, Q. B. Baker, and S. A. Al-Zboon, “A transfer learning with deep neural network approach for diabetic
retinopathy classification,” International Journal of Electrical and Computer Engineering (IJECE), vol. 11, no. 4, pp. 3492–3501,
Aug. 2021, doi: 10.11591/ijece.v11i4.pp3492-3501.
[20] H. Hakim, Z. Alhakeem, and S. Al-Darraji, “Goal location prediction based on deep learning using RGB-D camera,” Bulletin of
Electrical Engineering and Informatics (BEEI), vol. 10, no. 5, pp. 2811–2820, Oct. 2021, doi: 10.11591/eei.v10i5.3170.
[21] T. Ateeq et al., “Ensemble-classifiers-assisted detection of cerebral microbleeds in brain MRI,” Computers & Electrical
Engineering, vol. 69, pp. 768–781, Jul. 2018, doi: 10.1016/j.compeleceng.2018.02.021.
[22] S. B. Jadhav, “Convolutional neural networks for leaf image-based plant disease classification,” IAES International Journal of
Artificial Intelligence (IJ-AI), vol. 8, no. 4, pp. 328–341, Dec. 2019, doi: 10.11591/ijai.v8.i4.pp328-341.
[23] A. A. Alani and G. Cosma, “ArSL-CNN a convolutional neural network for Arabic sign language gesture recognition,”
Indonesian Journal of Electrical Engineering and Computer Science (IJEECS), vol. 22, no. 2, pp. 1096–1107, May 2021, doi:
10.11591/ijeecs.v22.i2.pp1096-1107.
[24] S. Lu, S.-H. Wang, and Y.-D. Zhang, “Detecting pathological brain via ResNet and randomized neural networks,” Heliyon, vol. 6,
no. 12, Dec. 2020, doi: 10.1016/j.heliyon.2020.e05625.
[25] M. S. AL-Huseiny and A. S. Sajit, “Transfer learning with GoogLeNet for detection of lung cancer,” Indonesian Journal of
Electrical Engineering and Computer Science (IJEECS), vol. 22, no. 2, pp. 1078–1086, May 2021, doi:
10.11591/ijeecs.v22.i2.pp1078-1086.
[26] P. C. Nissimagoudar, A. V. Nandi, A. Patil, and G. H. M., “AlertNet: Deep convolutional-recurrent neural network model for
driving alertness detection,” International Journal of Electrical and Computer Engineering (IJECE), vol. 11, no. 4, pp. 3529–
3538, Aug. 2021, doi: 10.11591/ijece.v11i4.pp3529-3538.
[27] M. A. Ihsan Aquil and W. H. Wan Ishak, “Evaluation of scratch and pre-trained convolutional neural networks for the
classification of Tomato plant diseases,” IAES International Journal of Artificial Intelligence (IJ-AI), vol. 10, no. 2, pp. 467–475,
Jun. 2021, doi: 10.11591/ijai.v10.i2.pp467-475.
[28] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in 2016 IEEE Conference on Computer
Vision and Pattern Recognition (CVPR), Jun. 2016, pp. 770–778, doi: 10.1109/CVPR.2016.90.
[29] X. Zhang, Y. Zou, and W. Shi, “Dilated convolution neural network with LeakyReLU for environmental sound classification,” in
2017 22nd International Conference on Digital Signal Processing (DSP), Aug. 2017, pp. 1–5, doi:

4407
10.1109/ICDSP.2017.8096153.
[30] Z. A. Khalaf, S. S. Hammadi, A. K. Mousa, H. M. Ali, H. R. Alnajar, and R. H. Mohsin, “Coronavirus Disease (COVID-19)
detection using deep features learning,” Scientific Reports, 2020, doi: 10.21203/rs.3.rs-60331/v1.
[31] A. H. H. Alasadi and A. K. H. Al-Saedi, “A method for microcalcifications detection in breast mammograms,” Journal of
Medical Systems, vol. 41, no. 4, Apr. 2017, doi: 10.1007/s10916-017-0714-7.
[32] M. Yildirim and A. Cinar, “Classification of Alzheimer’s disease MRI images with CNN based hybrid method,” Ingénierie des
systèmes d information, vol. 25, no. 4, pp. 413–418, Sep. 2020, doi: 10.18280/isi.250402.
BIOGRAPHIES OF AUTHORS
Faten Salim Hanoon received her BSc degree in Computer Science from
University of Basrah, Iraq. Currently, she is a Master's student in the Computer Information
Systems Department, College of Computer Science and Information Technology, University of
Basrah, Iraq. Currently. Her research interest is medical image processing and machine
learning. Her email account is fatin.salim369@gmail.com.
Abbas H. Hassin Alasadi is a Professor at the Computer Information Systems
Department, College of Computer Science and Information Technology, University of Basrah,
Iraq. He received his Ph.D. degree from the School of Engineering and Computer Science/
Harbin Institute of Technology, China. He spent more than ten years as Associate Professor at
different Universities abroad, the current position. His research interests include Medical
Image processing, Biometrics, Information retrieval, and Human-computer interaction. His
research work has been published in various international journals and conferences. Abbas is
an active reviewer in many journals of the areas of computer science and software engineering.
He is one of ACIT, UJCS, SIVP, and IJPRAI reviewer members. His email account is
abbas.hassin@uobasrah.edu.iq.

A modified residual network for detection and classification of Alzheimer’s disease

More Related Content

Similar to A modified residual network for detection and classification of Alzheimer’s disease (20)

More from IJECEIAES (20)

Recently uploaded (20)

A modified residual network for detection and classification of Alzheimer’s disease