Design and Development of
Efficient Soft Computing
Techniques for the
Alzheimer’s Disease
Classification
MADDALA JANAKIDEVI(21CSREP14)
Supervisor
Dr. CHERUKU RAMALINGA SWAMY
Assistant professor
Dept of CSE, NITW
Department of Computer Science and Engineering, NITW,
Warangal
DSC MEMBER
Dr. S Ravi Chandra, Head of the Dept
CSE(Chairman) 2
Dr. Ch Ramalinga Swamy, Asst. Professor, Dept of
CSE
Dr. P V Subba Reddy, Assistant professor, Dept of
CSE
Dr. Ilaiha Kavati, Assistant Professor, Department
Department
of CSE of Computer Science and Engineering, NITW,
Warangal
REGISTRATION DETAILS
Full Name Maddala Janakidevi
Registered Number 21CSREP14 3
Dr. Cheruku Ramalinga
Supervisor
Swamy
Date of
23rd December 2021
Registration
Type of registration Part-Time
Semester III
3
Department of Computer Science and Engineering, NITW,
Warangal
COURSE WORK
Course
Course Name Year
Type
Advanced Data Mining May 4
Regular
(CS5183) 2022
Soft Computing Techniques May
Regular
(CS5182) 2022
Self Dec
Deep Learning(CS5152)
Study 2022
Self Dec
Optimization in CS(CS5172)
Study 2022
4
Department of Computer Science and Engineering, NITW,
Warangal
CONTENTS
Introduction
Motivation
5
Related work
Research Objectives
Road Map
References
5
Department of Computer Science and Engineering, NITW,
Warangal
INTRODUCTION
Depression is one of the most prominent
and second major disabilities causing
mental illness affecting around 322 million 6
people worldwide.
Depression affects large number of people
across the world today and it is considered
as the global problem. It is a mood disorder
which can be detected using
electroencephalogram (EEG) signals.
Department of Computer Science and Engineering, NITW,
Warangal
METHODS USED FOR
CLASSIFICATION OF AD
Clinical assessment
Neuropsychological testing
7
Imaging studies
Biomarkers
Department of Computer Science and Engineering, NITW,
Warangal
MOTIVATION
• EEG signals respond to fluctuations of
affective states more sensitively and in real
time and thus can provide useful features of 8
emotional states.
• Different feature extraction methods such
as wavelet transform, (WT) and wavelet
packet transform (WPT) etc are available in
literature enhance the EEG signals via
features extraction. It is found that SVM, RF
and KNN types of classifiers are
successfully applied in this area for
Department of Computer Science and Engineering, NITW,
classification of emotions
Warangal
RESEARCH OBJECTIVES
Genetic Algorithm based Channel selection
for the Alzheimers Disease.
Design a Novel Framework based on 9
Machine Learning Model
Design a Novel Model based on Deep
Learning
Ensemble Model
Department of Computer Science and Engineering, NITW,
Warangal
Methods Framework
10
Methods Frame work Source :Fig 1(a) in [3]
Department of Computer Science and Engineering, NITW,
Warangal
CHANNEL MONTAGE
1
Anterior (Fp1, F3, Fz, Fp2, 1
F4)
Central (C3, Cz, C4)
Temporal/left (T3, T5, F7)
Temporal/right (T4, T6,
F8)
Posterior (P3, O1, Pz, P4,
O2)
Electrode clusters Source : Fig 1 in [4]
Department of Computer Science and Engineering, NITW,
Warangal
EEG RECORDING
1
2
Channel Montage Source : Fig 1(a) in [5]
ICA is widely used technique for artifact removal in EEG
signals. It is a statistical method that aims to separate mixed
signals into underlying independent componenet
Department of Computer Science and Engineering, NITW,
Warangal
A snapshot of the same signal
before and after being preprocessed
1
3
A snapshot of the same signal before and after being
preprocessed Source : Fig 1 in [1]
Department of Computer Science and Engineering, NITW,
Warangal
Feature Extraction
The most commonly extracted features for
EEG classification tasks is the Relative
Band Power (RBP) of the five frequency 14
bands of interest of the brain activity.
Delta: 0.5–4 Hz
Theta: 4–8 Hz
RBP = (Power in band)/(Total
Alpha: 8–13 Hz power)
Beta: 13–25 Hz
Gamma: 25–45 Hz
RBP Frequency bands in [4]
Department of Computer Science and Engineering, NITW,
Warangal
After removing
artifacts from the
data, the time
series information
is converted into
the frequency
domain using the
Fast Fourier
1
Transform (FFT). 5
This
transformation
enables the
calculation of band
powers for specific
Scalp heatmaps of PSD across 5 frequency bands ranges.
frequency Source : Fig
3 in [1]
Department of Computer Science and Engineering, NITW,
Warangal
AD,FTD, HC CLASSIFICATION
The dataset comprises data from 88
individuals who are categorized into 3
groups: 36 AD, 23 FTD and 29 HC, along
1
with Mini Mental State Examination(MMSE) 6
scores.
ICA is widely used technique for artifact
removal in EEG signals.
To extract spatial features from the data,
applied CNN to the matrix to capture
important features from the data.
The output from the CNN was then fed into
a LongofShort-Term
Department Memory
Computer Science (LSTM).
and Engineering, NITW,
Warangal
AD,FTD, HC CLASSIFICATION
The dataset comprises data from 88
individuals who are categorized into 3
groups: 36 AD, 23 FTD and 29 HC. 1
ICA is widely used technique for artifact 7
removal in EEG signals.
To extract spatial features from the data,
applied CNN to the matrix to capture
important features from the data.
The output from the CNN was then fed into
a Long Short-Term Memory (LSTM).
Department of Computer Science and Engineering, NITW,
Warangal
METHODOLOGY
18
METHODLOGY
Alzheimer's disease (AD) affects memory and
cognitive functions, progressing through three
stages: early, mild cognitive impairment (MCI),
19
and dementia.
Early diagnosis of AD is important for improving
patient survival, as traditional diagnostic
methods are difficult.
AI and deep learning methods are being
developed for faster and more accurate
detection of AD.
The model uses EEG signals from 19 different
brain areas for analysis.
Department of Computer Science and Engineering, NITW,
Warangal
METHODOLOGY Contd..
The collected EEG signals are preprocessed
using shallow autoencoders to remove noise
and finite impulse filters to remove artifacts.
2
0
The data is segmented using a sliding window
technique for better analysis.
Variational Mode Decomposition (VMD) and
Hilbert Transform are used to extract frequency
components and signal characteristics.
The extracted features are classified using a
neural network to detect early, mild, or severe
Alzheimer's disease, along with depression
levels.of Computer Science and Engineering, NITW,
Department
Warangal
Road Map
2
1
Department of Computer Science and Engineering, NITW,
Warangal
CONCLUSION
Alzheimer’s disease (AD) affects
memory and behavior in three stages.
EEG helps detect and monitor AD, with 22
machine learning aiding in
classification.
Reducing EEG channels improves
efficiency in diagnosis.
Machine learning methods like SVM and
CNN are effective for AD detection.
Department of Computer Science and Engineering, NITW,
Warangal
References
1. AlSharabi et al. (2022): Used discrete wavelet
transform and machine learning to process
EEG signals for Alzheimer's disease diagnosis.
[IEEE Access, 10, 89781-89797] 2
3
2. Xia et al. (2023): Developed a deep pyramid
CNN model based on EEG signals to diagnose
Alzheimer's disease. [Heliyon, 9(4)]
3. Li et al. (2021): Focused on feature extraction
and identification for Alzheimer's using multi-
channel EEG and latent factors. [IEEE
Transactions on Neural Systems and
Rehabilitation Engineering, 29, 1557-1567]
Department of Computer Science and Engineering, NITW,
Warangal
Department of Computer Science and Engineering, NITW,
Warangal
