Page 238

www.rsisinternational.org

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026

Neuro Imaging Assisted Cognitive Disorder Detection

Yegurla Vaagdevi, Vasa Haripriya

Department of Information Technology, Mahatma Gandhi Institute of Technology, Hyderabad,

Telangana, 500075, India

DOI:

https://doi.org/10.51583/IJLTEMAS.2026.150500023

Received: 27 April 2025; Accepted: 02 May 2026; Published: 25 May 2026

ABSTRACT

Cognitive disorders like Alzheimer’s disease and vascular dementia are global health concerns affecting millions

of people worldwide. It is important to diagnose these disorders at an early stage to effectively intervene and

provide treatment. This research proposes a complete computational framework for the detection of cognitive

disorders using sophisticated deep learning and machine learning techniques. The proposed system is based on

the combination of MRI analysis and CNN feature extraction VGG16 and Support Vector Machine (SVM)

classification tech niques. This proposed system can classify Alzheimer’s disease into four different stages: No

Impairment, Very Mild Impairment, Mild Impairment, and Moderate Impairment. Moreover, the pro posed

system can classify the presence or absence of vascular de mentia. This proposed system includes an explanation

mechanism using Grad-CAM to offer visual explanations for the predictions made bythe system. A Flask-based

web application is proposed to offer user-friendly access to the proposed diagnostic system with complete

reporting capabilities. Analysis of the proposed system shows promising results in terms of accuracy, precision,

recall, and F1-measure values. This work is a significant contribution to the new area of medical image analysis.

Keywords - Alzheimer’s Disease, Vascular Dementia, MRI, CNN, VGG16, Support Vector Machine, Grad-

CAM.

INTRODUCTION

One major issue around the world today involves problems with thinking and memory, such as vascular dementia

and Alzheimer’s disease. Over millions of people now live with some form of dementia- a number expected to

triple by 2050. While poor blood flow in the brain drives vascular dementia, another path lies behind most cases:

gradual shrinking of brain tissue, along with tangled proteins and swelling inside nerve cells. Alzheimer’s makes

up well over half of all dementias, marked clearly by clumps of amyloid-beta and twisted strands of tau. Unlike

sudden strokes, these changes creep slowly, eroding function piece by piece without clear warning signs early

on. Finding changes in the brain tied to thinking prob lems relies heavily on MRI scans. Because they show clear

images without needing surgery, these scans help spot issues Y. Vaagdevi Student, IT Department Mahatma

Gandhi Institute of Technology(MGIT) Hyderabad, Telangana, India Email: yvaagdevi csb223262@mgit.ac.in

sooner while tracking how conditions progress over time. Still, automated diagnosis tools become necessary

because reading vast brain scans by hand takes too much time, demands expert training, yet often differs between

readers. Tricky spots show up when spotting brain issues automatically- data gets too big, differences too small,

and rare cases get overlooked. Doctors expect answers they can trust, so systems must hit two targets: getting it

right, plus making their reasoning clear. From MRI scans, this research introduces a unified digital approach that

automatically sorts Alzheimer’s stages while spotting signs of vascular dementia. Instead of just listing f indings,

it shows active brain regions through visuals built into a browser tool. Built on layered algorithms, part smart

models and part transparent rules, the method aims to clarify predictions about how patients may progress. By

linking clear reasoning with complex pattern detection, it offers insights meant for real medical settings.

Accessible online, the setup targets remote care support, sharper forecasts, and better informed choices during

consultations. Visual feedback comes along each result, grounding conclusions in visible changes seen across

brain images.

Page 239

www.rsisinternational.org

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026

LITERATURE SURVEY

Alam and Latifi [1] investigated various neuroimaging modalities like MRI, PET, and CSF biomarkers for the

di agnosis of Alzheimer’s disease using various neuroimaging modalities. The study showed that deep learning

methods like CNN and CNN-LSTM produced very high accuracy of 99.95% and 99.92%, respectively.

Traditional methods like SVM are also successful for small datasets, but these methods are restricted by the high

computational cost and the need for large datasets. Alruily et al.

[2] built a smart system that spots Alzheimer’s through brain scans. Instead of relying on one method, they

stitched together traits from three known models- VGG16, MobileNet, along with InceptionResNetV2. Before

anything ran, the MRI pictures got cleaned up via resizing, balanced out by oversampling, then normalized and

stripped down to key details. Results landed near the top: 97.93% correct calls over all. Out of healthy cases,

nearly 98% were properly flagged as such. When sickness existed, it caught about 95.89% of those moments.

Right guesses among all positive reads settled at 95.94%. Awang et al. [3], have developed a model by using

DenseNet-201 architecture to classify Alzheimer’s disease by using MRI brain image. In this paper, AD5C

dataset is used, which contains 2381 classifications of MRI brain images in four classes. Various preprocessing

techniques are used to perform the proposed system. DenseNet-201 architecture is used to classify the feature

and stages of the disease. The proposed system is able to attain an accuracy of 98.24%, with high precision, F1-

score, and recall values. This shows that the proposed system is able to attain high accuracy in the diagnosis of

Alzheimer’s disease by using transfer learning and DenseNet-201 architecture. Tufail et al. [4] introduced a deep

learning model for the classification of vascular dementia from MRI and rs-fMRI brain images. The authors

have employed different transfer learning models such as DenseNet121, VGG16, VGG19, In ceptionResNetV2

for the detection of the disease. The authors have processed and balanced the data using different sampling

methods. The authors have split the brain images into training, validation, and testing sets (80%, 10%, 10%).

Activation functions such as ReLU and Leaky ReLU have also been used to overcome the vanishing gradient

problem.

Among all the models, DenseNet121 has been found to perform better for the classification problem. The

proposed model has been found to achieve 84.67% accuracy for multi-class classification for the detection of

vascular dementia using MRI brain scans. In the review by Morgan and McAuley [5] on the patho biology, risk

factors, and emerging research perspectives in vascular dementia (VaD), it is stated that VaD is the 2nd most

common form of dementia, usually resulting from cerebrovas cular abnormalities in brain blood flow. It is caused

by various risk factors like hypertension, obesity, diabetes, smoking, di etary habits, and physical inactivity. It

is also caused by various pathological mechanisms like oxidative stress, inflammation, and breakdown of the

blood–brain barrier. It is mentioned that emerging research perspectives include epigenetics, mi crobiome

interactions, and various potential compounds like metformin, rapamycin, and resveratrol. Sampath and Baskar

proposed the Fly Optimized Densely Connected CNN (FODC-CNN) [6], which is a method for the prediction

of Alzheimer’s disease from the MRI images of the brain. The method employs adaptive histogram enhancement

and weighted median filters for the enhancement of the quality of MRI images and the removal of noise from

the images. The extraction of features is done by the convolutional clustering method, and the areas affected by

Alzheimer’s disease are detected.

The parameters of the CNN are optimized by the fly optimization algorithm for the improvement of the classifica

tion accuracy. The results achieved by the proposed method on the Kaggle MRI dataset with more than 6400

images proved the efficiency of the proposed method for the diagnosis of AD. Chouliaras and O’Brien [7]

conducted a study that reviewed the role of neuroimaging techniques and the differential di agnosis of dementia.

The study highlights the different neu roimaging techniques such as PET, MRI, and SPECT scans that are used

for the detection of abnormalities in the brain that are associated with Alzheimer’s dementia, vascular dementia,

and Lewy body dementia. The study concludes that the use of neuroimaging and computational techniques can

enhance the accuracy of dementia diagnosis. A deep learning framework was proposed by Nithya et al. [8] based

on ResNet-50 for early classification of Alzheimer’s disease using MRI brain image data. In this proposed frame

work for image classification, different preprocessing tech niques such as BADF and CLAHE are used for image

enhancement. In this proposed framework for image classi f ication, K-means clustering techniques are used for

image segmentation. In this proposed framework for image classi f ication, the ResNet-50 model with shortcut

Page 240

www.rsisinternational.org

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026

connections is used to extract features from brain image data and to classify normal and Alzheimer’s disease

image data. In this proposed framework for image classification, an accuracy of 95% is achieved in classification

with low loss using ADNI image data. In the paper published by Yao et al. [9], a brief idea about the use of Ai

in the diagnosis of Alzheimer’s disease by analyzing the MRI image data was provided. The various models of

Ml such as Attention Networks, Auto-Encoders, CNN, GANs, Deep Belief Networks, etc. have been discussed

for the identification of the structural changes that take place in the brain due to Hippocampal Atrophy and

Cortical De generation. The various advanced techniques such as Transfer Learning, Contrastive Learning, Meta

Learning, Federated Learning, Multimodal Data, etc. have been discussed for the early diagnosis of Alzheimer’s

disease. A model for the classification of Alzheimer’s disease using MRI brain images has been proposed by

Sharma et al. [10] using CNN-MobileNetV2. The authors have utilized differ ent preprocessing methods such

as median filtering, FFT, and DWT for enhancing the quality of the image. Feature extraction methods such as

HOG, GLCM, and PSO-SURF have also been utilized.

The extracted features have been employed for the classification of images into Normal Control, Mild Cognitive

Impairment, and Alzheimer’s Disease. The proposed model has achieved 97% accuracy on the Kaggle

MRIdataset compared to other models such as VGGNet, CNN and MobileNetV2. Lampe et al. [11] has

developed a ML model using volumet ric MRI imaging to predict and distinguish multiple dementia syndromes.

The models have been trained using T1-weighted MRI scans of 426 patients and 51 healthy participants. The

binary models have demonstrated 71% to 95% accuracy in the models, and the multi-syndrome classifier

demonstrated an accuracy of 47.4% in distinguishing seven different dementia syndromes. The models have

demonstrated higher accuracy in models with distinct brain atrophy patterns like svPPA, bvFTD, and PSP.

Sharma et al. [12] proposed a model using deep learning with CNN for detecting Alzheimer’s disease by

analyzing the MRI scans of brain images. The proposed model was trained on two different datasets of MRI

scans, one containing 6400 images and the other containing 6330 images. The images were categorized into

various stages, such as Non-Demented, Very Mild Demented, Mild Demented, and Moderate De mented.

The VGG16 model was employed to extract features from MRI scans, and a neural network was employed to

classify images. The proposed model attained an accuracy of 90.4%, precision of 0.905, recall of 0.904, F1 score

of 0.904, and area under the curve of 0.969 on the first dataset. The proposed model validated that deep learning

using VGG16 is efficient for diagnosing early-stage Alzheimer’s disease. Lee et al. [13] proposed the brain

imaging abnormalities in mixed Alzheimer’s disease and subcortical vascular dementia using advanced MRI

techniques such as T2-FLAIR MRI, DTI, QSM, and R2 relaxation.

They analyzed the data from 17 participants (5 AD, 5 SVaD, 7 mixed dementia) and observed that the patients

suffering from mixed dementia have higher white matter signal abnormalities, especially in the frontal lobes,

and lower R2 values compared to the patients suffering from Ad and subcortical Vd.

It has been observed that the mixed dementia shows imaging characteristics of both Alzheimer’s and vascular

brain damage, which indicates that the MRI biomarkers can help in the differential diagnosis of the different

types of dementia. Alizadeh et al. [14] in the study have proposed a deep learning method based on the analysis

of the rs-fMRI data for the classification of Alzheimer’s disease, Mild Cognitive Impairment, and normal

controls. In the proposed method, the BOLD signal in 39 brain regions, according to the MSDL atlas, has been

used. A 1D CNN has been proposed in the study to classify Ad and normal controls.

The dataset, which consists of 15 Alzheimer’s disease patients, 17 Mild Cognitive Impairment patients, and 10

normal controls, has been used in the proposed method. The signal augmentation has been used in the study to

increase the size of the dataset. The experimental results show that the overall accuracy of the proposed method

is 68.5%, the precision is 0.681 and the F1 score is 0.663. In the paper presented by Basheer et al. [15], a novel

model named MCapNet, which is a modified deep neural network using capsule network architecture, has been

introduced for predicting dementia using the OASIS MRI dataset. The dataset considered has 373 samples with

15 clinical and imaging features. The dataset has been classified into demented and non-demented. The EDA

analysis has shown that age and gender play a significant role in predicting dementia. The model introduced,

i.e., MCapNet, has shown 92.39% accuracy compared to other Ml classifiers such as Decision Trees, Random

Forest, SVM, Gradient Boosting, etc.

Page 241

www.rsisinternational.org

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026

METHODOLOGY

Proposed Architecture

The proposed system will employ a complete system for the automated detection of cognitive disorders. The

system’s architecture will comprise five major components:

Preprocessing and Image Acquisition

MRI images ac quisition from clinical settings followed by normalization to a standard size of 224x224 pixels

with histogram equalization for optimal contrast.

Feature Extraction

Convolutional neural networks pre trained are used as the feature extractor to produce compact and fixed-sized

feature vectors by processing representations from MRI images. The proposed system uses the VGG16 model

as the backbone, whereas others are tested as the baseline backbones. The classification head is excluded from

each backbone, making each one a deep feature extractor. This is done by performing global average pooling on

the feature vector maps of the last convolutional layer and passing them to classical machine learning models.

Classification

Features will be classified into disease types by employing a SVM with a radial basis function kernel.

Explainability and Visualization

Grad-CAM will be used to generate heatmaps for the input images to highlight the most influential areas of the

images for the prediction task.

Reporting and User Interface

A web interface will be developed to enable easy usage of the system.

Data Preparation and Preprocessing

Dataset Structure

The system uses two different data sets with the following structure:

Alzheimer’s MRI Dataset - Includes MRI images that fall under 4 different classes based on the stages of Ad:

 No Dementia- Control baseline

 Very Mild Dementia- Early changes

 Mild Dementia- Mild Cognitive Impairment stage

 Moderate Dementia- Alzheimer’s diagnosis

Vascular Dementia - Dataset Binary classification dataset with the following two classes:

 Present- Presence of vascular pathology

 Absent– Absence of vascular disease indicators

Page 242

www.rsisinternational.org

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026

Image Preprocessing Pipeline

Standardized processing for each image obtained from the MRI scan:

Loading: Images are loaded in grayscale format

Histogram Equalization: Equalization is performed for image contrast enhancement

Resizing: Images are resized to 224x224 pixels for CNN input

Normalization: Normalization of pixel intensities in the range of [0,1] and pre-processing by backbone’s own

func tions

Feature Extraction Architecture

The extraction of features is done using the pretrained deep learning networks such as VGG16, VGG19,

ResNet50, and MobileNetV2. In the proposed method, the VGG16 network is used as the primary feature

extractor, while other networks are considered as baseline methods. The last classification layer is discarded,

and global average pooling of the extracted convoluted features yields a fixed-size feature vector. Key features

of the model include:

Architecture

The efficiency and effectiveness of this model are ensured through the use of depthwise separable convolutional

layers and inverted residual blocks in its architecture. It comprises a total of 28 convolutional layers, ensuring a

decrease in complexity while increasing efficiency in feature extraction. Furthermore, it utilizes pre-trained

weights from the ImageNet dataset, ensuring a boost in learning potential through transfer learning.

Output Specification

Global average pooling of the last convolutional feature maps to generate fixed-size 1280-dimensional feature

vectors while discarding spatial resolution while preserving semantic content.

Explaining Models Using Grad-CAM

To improve clinical interpretability, this system also pro vides visual explanations in the form of Grad-CAM

images:

Mechanism

1. Compute gradients of class scores with respect to activa tion in the convolutional layer

2. Weight activation maps using gradient magnitude

3. Produce high-resolution heatmaps using bilinear interpo lation.

4. Overlays heatmaps on original MRI images using JET colormap (red = high relevance)

Clinical Utility

The heatmaps help doctors verify the decision-making process of the model and pinpoint patholog ical features.

Training and Validating Models

Data Splitting

Training and evaluation process follows the stratified cross-validation approach:

Page 243

www.rsisinternational.org

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026

1. Training Set: 80

2. Test Set: 20

3. Stratified Cross Validation: This ensures that the class distribution is preserved.

4. Random Seed: 42

Training Process

During training, disease classification models go through several stages laid out here. Starting off, image features

come from MRI scans by way of the VGG16 system. Following that stage, those captured traits feed into a

Support Vector Machine, shaping how the model learns to sort conditions. Once learning finishes, real testing

begins using unseen data to check how well it works. From there, scores get worked out to show just how

accurate things turned out. To wrap up, both the trained setup and its results get stored ready for later deployment

down the line.

Web Application Architecture

A web interface for the end-user to interact with the diag nostic system is implemented by creating a web

application with the Flask web application framework:

Backend Components

1. Flask web application framework for handling HTTP requests.

2. SQLite database for patient record management.

3. Image processing pipeline for handling uploaded images containing MRI scan results.

4. Model inference engine with caching mechanisms imple mented.

5. ReportLab PDF generation for creating reports for the patient

Frontend Features

1. Drag-and-drop interface for image upload.

2. Real-time prediction with confidence level visualization.

3. Interactive heatmap display.

4. Patient record management history.

5. PDF report download.

RESULTS AND DISCUSSION

Measuring the model Performance

Classification Metrics

Classification metrics to measure the effectiveness of classifications were applied to the held-out test set.

Accuracy

The most basic measure of effectiveness is the accuracy of the classification system as a measure of the

Page 244

www.rsisinternational.org

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026

proportion of samples that were correctly predicted. A total accuracy of a classification system in this context

can be expressed as:

Accuracy = (TP + TN) / (TP + TN + FP + FN)

Precision

The precision metric represents the reliability of the positive prediction. The total precision of a classification

system can be represented as:

Precision = TP / (TP + FP)

Recall

Recall metric represents the ability of the model to identify the true positives. Total recall of the model can be

expressed as:

Recall = TP / (TP + FN)

F1-Score

The F1-Score is a combined measure of recall and precision. The F1-Score of the model can be expressed as:

F1 Score = 2 × (Precision × Recall) / (Precision + Recall)

Where, TP = True Positives, TN = True Negatives, FP = False Positives, FN = False Negatives.

Performance Results

The results suggest that the choice of using the VGG16 as the feature extractor and the SVM as the classifier

offers adequate diagnostic capability to support clinical decision support systems.

Evaluation of the Confusion Matrices

The evaluation of the confusion matrices provided additional insight into the behavior patterns of the classifiers.

Alzheimer’s Classifier

The confusion matrix has strong diagonal dominance; there fore, the correct classification of Ad cases was high.

The majority of misclassifications occurred in adjacent stages of the disease, which is clinically acceptable due

to the gradual, rather than discrete, nature of cognitive impairment and the decline along the disease continuum.

Page 245

www.rsisinternational.org

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026

Vascular’s Classifier

The binary classification produced a clear separation be tween classes; therefore, the potential for false-positive

results was low, minimizing the need for follow-up clinical eval uations, as well as increasing the overall

reliability of the diagnosis.

Performance Comparision Analysis

Table 1. Alzheimer’s Diseease Comparison

Model

Accuracy

Precision

Recall

F1 Score

VGG16 + SVM

96.58%

96.62%

96.58%

96.59%

MobileNetV2 + SVM

93.42%

93.41%

93.42%

93.41%

VGG19 + SVM

95.67%

95.73%

95.67%

95.68%

ResNet50 + SVM

94.08%

94.11%

94.08%

94.09%

Table 2. Alzheimer’s Diseease Comparison

Model

Accuracy

Precision

Recall

F1 Score

VGG16 + SVM

84.20%

MobileNetV2 + SVM

78.60%

VGG19 + SVM

80.10%

ResNet50 + SVM

79.40%

Explainability Analysis Using Grad-CAM

In order to give insight into the interpretability of deep learning models, Grad CAM was used for visualization

pur poses. When the heatmaps were generated, they demonstrated the presence of activation patterns that were

Page 246

www.rsisinternational.org

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026

clinically relevant. There are areas of the hippocampus and medial temporal regions that exhibited a very high

level of activation, which mirrors the pathology of individuals with early Alzheimer’s disease. Those with

moderate impairments demonstrated a much larger area of cortical activation than those with either high or low

impairments. The areas of periventricular white matter that were highlighted by the vascular dementia predic

tions matched what is known about the vascular pathology of vascular dementia. These results validate that the

deep learning model learns structural features that are relevant to medicine rather than artifacts that are unique

to the dataset used for training.

DISCUSSION

The proposed framework provided a performance level comparable to other existing deep learning methods for

neu roimaging analysis while maintaining computational efficiency and interpretability. Effectiveness of

Architecture Due to the following variables, VGG16 was the architecture of choice for the proposed framework:

Is low in computational complexity; Has high speed of inference; and Is effective in transfer learning for

neuroimaging analysis. By using the SVM classifier, the robustness of the frame work increased when training

using a small number of medical datasets and allowed for high-quality calibrated probability estimates that would

allow for clinical interpretation. Clinical Relevance The suggested framework would act as a medical decision

support tool for clinicians. It helps detecting early signs of cognitive impairment and determining the phase and

MRI scans to identify vascular dementia. In this way, the automated system alleviates the task of manual MRI

interpretation and enhances diagnostic consistency. The web-based interface, furthermore, allows for remote

access, enabling telemedicine applications in rural or resource-limited healthcare settings.

CONCLUSION

The current research is proposing a wide framework that can be used for the automatic detection of cognitive

disorders with the help of neuro-imaging data using deep learning techniques as well as interpretable machine

learning techniques. This proposed system utilizes VGG16 and Support Vector Machine classification

techniques to guarantee the reliability of the proposed framework while incorporating Grad-CAM visual izations

to provide interpretability to the proposed system. A user-friendly web application has also been proposed to

provide clinicians with access to the proposed system without requiring technical expertise. This proposed

system has been evaluated using extensive performance metrics, confusion ma trix analysis, and explainability

analysis to guarantee reliability and relevance. Future research directions include the incorporation of various

Page 247

www.rsisinternational.org

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026

imaging modalities like functional MRI, PET scans, and diffusion tensor imaging to incorporate various types

of pathological information. Longitudinal imaging informa tion can also be used to analyze the disease

progression or future cognitive decline. Future enhancements can also include demographic modeling,

incorporation of biomarkers with genetic information and cerebrospinal fluid information, and making the

system more robust to adversarial variations that often occur in a clinical setting. Clinical validation through

prospective studies and regulatory approval from the FDA or CE mark will be necessary for the system to be

used clinically.

REFERENCES

1. Alam, MD Minul, and Shahram Latifi. ”A Systematic Review of Tech niques for Early-Stage

Alzheimer’s Disease Diagnosis Using Machine Learning and Deep Learning.” Journal of Data Science

and Intelligent Systems (2025).

https://ojs.bonviewpress.com/index.php/jdsis/article/view/5037

2. Alizadeh, Farzad, et al. ”Differential diagnosis among alzheimer’s disease, mild cognitive impairment,

and normal subjects using resting state fMRI data extracted from multi subject dictionary learning atlas.”

Frontiers in Biomedical Technologies 9.4 (2022): 297-306.

https://publish.kne-

publishing.com/index.php/fbt/article/view/10423

3. Alruily, Meshrif, et al. ”Ensemble deep learning for Alzheimer’s disease diagnosis using MRI:

Integrating features from VGG16, MobileNet, and InceptionResNetV2 models.” PloS one 20.4 (2025):

e0318620. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0318620

4. AV, AMBILI, DAS KUMAR, and DR LATIP. ”CNN-Mobilenetv2 Deep Learning-Based Alzheimer’s

Disease Prediction and Classifica tion.” Journal of Theoretical and Applied Information Technology

101.9 (2023). http://www.jatit.org/volumes/Vol101No9/32Vol101No9.pdf

5. Awang, Mohd Khalid, et al. ”Classification of Alzheimer disease using DenseNet-201 based on deep

transfer learning technique.” Plos one 19.9 (2024): e0304995.

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0304995

6. Basheer, Shakila, Surbhi Bhatia, and Sapiah Binti Sakri. ”Computational modeling of dementia

prediction using deep neural network: analysis on OASIS dataset.” IEEE access 9 (2021): 42449-42462.

https://ieeexplore.ieee.org/abstract/document/9380278/

7. Chouliaras, Leonidas, and John T. O’Brien. ”The use of neuroimaging techniques in the early and

differential diagnosis of dementia.” Molec ular Psychiatry 28.10 (2023): 4084-4097.

8. Lampe, Leonie, et al. ”Multiclass prediction of different dementia syn dromes based on multi-centric

volumetric MRI imaging.” NeuroImage: Clinical 37 (2023): 103320.

https://www.sciencedirect.com/science/article/pii/S2213158223000098

9. Lee, Hyunwoo, et al. ”Brain imaging abnormalities in mixed Alzheimer’s and subcortical vascular

dementia.” Canadian Journal of Neurological Sciences 50.4 (2023): 515-528.

https://www.cambridge.org/core/journals/canadian-journal-of-neurological-sciences/article/brain-

imaging-abnormalities-in-mixed-alzheimers-and-subcortical-vascular-

dementia/62C806DD19FE77608FD91159799AA445

10. Morgan, Amy Elizabeth, and Mark Tom´as Mc Auley. ”Vascular de mentia: from pathobiology to

emerging perspectives.” Ageing research reviews 96 (2024): 102278.

https://www.sciencedirect.com/science/article/pii/S1568163724000965

11. Nithya, V. P., N. Mohanasundaram, and R. Santhosh. ”An early detection and classification of

Alzheimer’s disease framework based on ResNet 50.” Current Medical Imaging 20.1 (2024):

e250823220361.

https://www.benthamdirect.com/content/journals/cmir/10.2174/1573405620666230825113344

12. Sampath, R., and M. Baskar. ”Alzheimer’s Disease Prediction Using Fly-Optimized Densely Connected

Convolution Neural Networks Based on MRI Images.” The Journal of Prevention of Alzheimer’s

Disease 11.4 (2024): 1106-1121.

https://www.sciencedirect.com/science/article/pii/S2274580724000372

13. Sharma, Shagun, et al. ”A deep learning based convolutional neural network model with VGG16 feature

extractor for the detection of Alzheimer Disease using MRI scans.” Measurement: Sensors 24 (2022):

100506. https://www.sciencedirect.com/science/article/pii/S2665917422001404

14. Tufail, Hina, et al. ”Classification of vascular dementia on magnetic res onance imaging using deep

Page 248

www.rsisinternational.org

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026

learning architectures.” Intelligent Systems with Applications 22 (2024): 200388.

https://www.sciencedirect.com/science/article/pii/S2667305324000632

15. Yao, Zhaomin, et al. ”Artificial intelligence-based diagnosis of Alzheimer’s disease with brain MRI

images.” European Journal of Radiology 165 (2023): 110934.

https://www.sciencedirect.com/science/article/pii/S0720048X23002486