ADVANCING DEMENTIA DETECTION

AND CLASSIFICATION

• UNDER THE SUPERVISION OF:

• DR. JOLLY PARIKH

BHOOMI KHURANA: 0211502721

ALI ZYADI: 01711502721
RISHABH GUPTA: 04611502721
PRIYANSH KASHYAP:04711502721
1 Department Name, BVCOE, New Delhi
Topic Name: , Mentor Name:
 CONTENTS

 INTRODUCTION
 PROBLEM STATEMENT
 PROPOSED METHODOLOGY
 OBJECTIVES
 LITRATURE REVIEW
 IMPLEMENTATION
 REFERENCES
Department Name, BVCOE, New Delhi
2 Topic Name: , Mentor Name:
 INTRODUCTION

 Dementia, with Alzheimer's disease (AD) as its most prevalent form, is an escalating
global health concern. Timely detection and classification of dementia play pivotal
roles in patient care and intervention strategies. The challenge lies not only in
identifying early symptoms but also in providing accessible and accurate diagnostic
tools [1]. In this context, our research endeavors to harness the power of deep
learning and machine learning techniques to create a robust system for dementia
detection.

 Our research sets out with a clear purpose: to develop an innovative, accessible, and
highly accurate system for dementia detection. By achieving this objective, we hope
to make substantial strides in addressing the global challenge of dementia, ultimately
contributing to the well-being of individuals and society at large.
3
 PROBLEM STATEMENT

• Detecting dementia, especially Alzheimer's disease (AD), in its early

stages is a critical challenge. There's also a lack of user-friendly tools
that integrate advanced computational models to analyze MRI scans.
Combining diverse datasets securely, like MRI scans, PET scans is
complex. It utilizes deep learning and machine learning to offer precise,
early-stage dementia detection and classification.

4
 PROPOSED METHODOLOGY

Model Exploration
LITRATURE REVIEW • Experimenting with
• Learning approach DATA PREPROCESSING
• Replicating studies
• Data Cleaning & Exploration different Models
• Normalizing & Augmentation • Finding techniques and
possible model

WEBSITE HYBRID MODEL

DEVELOPMENT DEVELOPMENT
• Front-End MODEL TRAINING AND
• Expanding Dataset DEVELOPMENT
Development
• Back-End • Training Individual models • Implementing Model
• Model Integration • Hyperparameter Tuning
Development
• Model Integration • Fine-Tuning • Validaton
 OBJECTIVES
Objectives:

• Exploring recent studies.

• Testing out various models and finding a suitable one for the data.
• Applying Neural Networks on dataset for detection of dementia.
• building a user-friendly website for MRI scans and vital data input.
 LITRATURE REVIEW

AUTHOR METHODOL DESCRIPTION RESULT LIMITATIO

OGY N
[1] F. Zhang, Z. Li, Stacked In their work, stacked autoencoder, a Binary classification While Binary class
B. Zhang, H. Autoencoder, regression layer of softmax, accuracy= 88.58% accuracy was pretty
Du, B. Wang, CNN and minimal labelled learning samples utilis- high, the4 class
X. Zhang were used and CNN was applied on Ing MRI and PET accuracy had poor
ADNI dataset and PET Scans for images, and 4-class results.
Biinary and 4-class Classification. classification
accuracy was
47.42%

[2] M.P. Bhatkoti KSA classification k-Sparse autoencoder (KSA) was Enhanced pair Model complexity
Pushkar used To find degraded brain regions. predictions with 100 was very high and
Experiments were conducted using classifier there is overfitting
150 images for MRI scans as well as combinations issue with degraded
CSF and PET images from the ADNI had accuracy = brain region study.
study. 83.143%, compared
to 71.327% for 50
combinations

[3] F. CHIYU, A. 3D-CNN A 3D-CNN framework for AD diag- Classifying AD Model was unable
ELAZAB, Y. framework, FSBi- nosis that utilized both MRI and PET from NC, pMCI from to diagnose sMCI
PENG, T. LSTM images, To improve the results of their NC, and sMCI from with good accuracy
WANG, F. method, they used FSBi-LSTM to NC had diagnosis which limits the real
ZHOU completely stack bidirectional long accuracy results of life application of
short- term memory on the concealed 94.82%, 86.36%, study.
spatial information from deep and 65.35%,
feature maps on ADNI dataset. respectively.
 LITRATURE REVIEW

AUTHOR METHODO DESCRIPTION RESULT LIMITATI

LOGY ON
[4] J. Islam, Y. Zhang 3 CNNs Multi- Their technique is made up of The accu- validation
model three deep CNNs, each with a racy, precision, recall, database had
little different structure. and F1-score were only 416 sMRI
93.18%, 94%, 93%, data.
and 92%, respectively.

[5] M. Liu, F. Li, H. CNN, 3D A deep CNN model was used for Their method had
Yan, K. Wang, Y. DenseNet hippocampal segmentation. accuracy of 88.9%,
Ma, L. Shen, M. Xu, Then, using the segmented (ROC) had accuracy
A.D.N hippocampus a 3D DenseNet 76.2%. and an AUC of
was created to learn distinct 77.5% for detecting MCI
image features. vs. NC subjects.

[6] S. Neffati, K. Ben DKPCA, SVM They downsized the DKPCA was tested on Model focused
Abdellafou, I. Jaffel, dimensionality of the data using the OASIS dataset and more on
O. Taouali, K. DKPCA, and then applying a results were satisfactory preprocessing
Bouzrara multiclass SVM for classification compared to other methodology and
on the reduced-dimensional conventional less research
data. approaches. was done in
classification.

[7] T.A. Tuan, T.B. CNN, GNN, First they did tissuesegmentation On AD-86 and AD-126
Pham, J.Y. Kim, XGBoost, SVM which integrates a CNN module datasets, they achieved
J.M.R. Tavares and a Gaussian Mixture Model a Dice of 0.96 for
(GMM).Then, a hybrid classifica- segmentation and
tion model is proposed that classification accuracies
combines both extreme gradient o 0.88 and 0.80,
boosting (XGBoost) and SVM to respectively. AD-126
classify AD depending on group’s classification
the segmented tissues. accuracy was
lower than AD-86.
 LITRATURE REVIEW

AUTHOR METHOD DESCRIPTION RESULT LIMITATI

OLOGY ON
[8] S. Sarraf, G. CNN (LeNet) A functional MRI (fMRI) study 96.86% Can only
Tofighi was proposed for the classification accuracy for AD classify AD and
of AD. The technique vs NC NC, but not
is based on the LeNet DL Classification. MCI.
architecture and utilizes transfer
Learning.

[9] Z. Cui, Z. Gao, J. CNN (Inception To improve the focus accuracy Accuracy of
Leng, T. Zhang, P. V3) Inception V3 three Inception blocks 85.7% for NC vs
Quan, W. Zhao were used. Better Inception V3 MCI
mannequin classified patients Classification.
With NC and MCI effectively.

[10] E. Jabason, M.O. Ensemble of To categorize MRI images Ensemble

Ahmad, M. Swamy CNNs(DenseNet and identify patients with AD, MCI, method had a
121,DenseNet16 and NC OASIS data-set was used to 95.23%
9, DenseNet201, evaluate this ensemble using Accuracy.
and ResNet50) structural MRI.

[11] M. Shahbaz, S. Ali, KNN, DT, GLM Their study compared six different GLM can The accuracy
A. Guergachi, A. ML and data mining algorithms effectively was not very
Niazi, A. Umer including kNN, Decision Tree (DT), diagnose the high, and the
rule stages of AD work represents
inducton, Naive Bayes, generalized with accuracy of a comparative
linear model (GLM), and DL 88.24% study of ML
algorithms to find most distinguishing techniques.
feature for AD staging. All six
classifiers were subjected to a
10-fold cross-validation procedure.
 IMPLEMENTATION OVERVIEW
1. Explored diverse Kaggle brain image datasets with classes like 'Moderate Demented,'
'Non Demented,' 'Mild Demented,' and 'Very Mild Demented.’

• 2. Implemented a CNN for image classification using TensorFlow and Keras, Conducted
rigorous exploratory data analysis, visualizing random images and scrutinizing
dimensions.

• 3. Preprocessed data with normalization and caching for efficient processing, Compiled
the CNN with the Adam optimizer and performed insightful training with visualizations.

• 4. Applied data augmentation techniques to enhance model generalization and found

validation accuracy and plotted graph.

• 5. Built a front-end for a web application for dementia detection and laid the foundation
for back-end development.
 DATASET OVERVIEW

In this project, the foundation of our exploration is a robust dataset

obtained from Kaggle, showcasing a rich array of brain images. The
dataset encapsulates distinct classes as ‘Non Demented', 'Mild Demented',
'Moderate Demented', and 'Very Mild Demented'.
 Convolutional Neural Networks
(CNNs)
Definition: Convolutional Neural Networks (CNNs) are a class of deep
neural networks designed for processing structured grid data, such as
images.

Key Components:

• Convolutional Layers: These layers apply filters to detect patterns and

features in input images.
• Pooling Layers: Used for downsampling and reducing spatial
dimensions, focusing on essential features.
• Flatten Layer: Converts 2D feature maps to a 1D array for input to
dense layers.
• Dense Layers: Fully connected layers for classification or regression
tasks.
 Application to Our Project
Dataset: Leveraged Kaggle datasets containing diverse brain images
('Non_Demented', 'Mild_Demented', etc.)

Model Architecture:
•Input Layer: Rescales pixel values to [0, 1] for normalization.
•Convolutional Layers: Extract features via convolution and activation
functions (ReLU).
•Pooling Layers: Downsample spatial dimensions.

• Flatten Layer:
Converts 2D feature
maps to a 1D array
• Dense Layers:
Concludes with dense
layers for
classification.
 PREPROCESSING
• Checked class-wise distribution of images in the dataset.

• Visualized random images from each class.

• Inspected image dimensions and characteristics.

• Prepared image datasets for training and validation.

• Applied data normalization and caching for efficient processing.

DATA AUGUMENTATION AND DEFINING CNN

• Implemented data
augmentation techniques
(horizontal flip, rotation, zoom)
to enhance model
generalization.

• Constructed a CNN model with

multiple convolutional layers
• Compiled the model using the
Adam optimizer and sparse
categorical crossentropy loss.
Model Validation Results

Visual Representation:
Graphical presentation showcasing
the validation accuracy and loss
curves over epochs.

Validation Accuracy:
Achieved an impressive validation
accuracy of 98.68%.
Validation Loss:
Attained a low validation loss of
0.0411.
 REFERENCES
[1] F. Zhang, Z. Li, B. Zhang, H. Du, B. Wang, X. Zhang, Multi-modal deep learning
model for auxiliary diagnosis of alzheimer’s disease, Neurocomputing 361 (2019) 185–
195.
[2] M.P. Bhatkoti Pushkar, Early diagnosis of alzheimer’s disease: A multi-class deep
learning framework with modified k-sparse autoencoder classification.
[3] F. CHIYU, A. ELAZAB, Y. PENG, T. WANG, F. ZHOU, Deep learning framework
for alzheimer’s disease diagnosis via 3d-cnn and fsbi-lstm.
[4] J. Islam, Y. Zhang, Brain mri analysis for alzheimer’s disease diagnosis using an
ensemble system of deep convolutional neural networks, Brain informatics 5 (2) (2018)
1–14
[5] M. Liu, F. Li, H. Yan, K. Wang, Y. Ma, L. Shen, M. Xu, A.D.N. Initiative, et al, A
multi-model deep convolutional neural network for automatic hippocampus
segmentation and classification in alzheimer’s disease, NeuroImage 208 (2020)
116459.
[6] S. Neffati, K. Ben Abdellafou, I. Jaffel, O. Taouali, K. Bouzrara, An improved
machine learning technique based on downsized kpca for alzheimer’s disease
17 classification,Department
Int. J. Imaging Syst. Technol.
Name, BVCOE, 29 (2) (2019) 121–131
New Delhi
Topic Name: , Mentor Name:
 REFERENCES
[7] S. Sarraf, G. Tofighi, Classification of alzheimer’s disease using fmri data and deep
learning convolutional neural networks, arXiv preprint arXiv:1603.08631.

[8] Z. Cui, Z. Gao, J. Leng, T. Zhang, P. Quan, W. Zhao, Alzheimer’s disease diagnosis
using enhanced inception network based on brain magnetic resonance image, in: 2019
IEEE International Conference on Bioinformatics and Biomedicine (BIBM), IEEE,
2019, pp. 2324–2330.

[9] E. Jabason, M.O. Ahmad, M. Swamy, Classification of alzheimer’s disease from mri
data using an ensemble of hybrid deep convolutional neural networks, 2019 IEEE 62 nd
International Midwest Symposium on Circuits and Systems
(MWSCAS), IEEE, 2019, pp. 481–484.

[10] M. Shahbaz, S. Ali, A. Guergachi, A. Niazi, A. Umer, Classification of alzheimer’s

disease using machine learning techniques, DATA (2019) 296–303.

[11] T.A. Tuan, T.B. Pham, J.Y. Kim, J.M.R. Tavares, Alzheimer’s diagnosis using deep
learning in segmenting and classifying 3d brain mr images, Int. J. Neurosci. (2020) 1–
10.
THANK
YOU