Dementia Classification
SVECW |Department of Artificial Intelligence |
Project Review-1
Project Batch Number: "06" -
Team Members: Jahnavi K (21B01A5449)
Nandini D (21B01A5431)
Manvitha S (22B05A5406)
Veena G (21B01A5438)
Faculty Guide: Mr. V. Harinadh, Assistant Professor,
Artificial Intelligence, SVECW
2.
Motivation
• Global Impact:Dementia is a growing concern worldwide,
necessitating early and accurate diagnosis for better healthcare
outcomes.
• Data-Driven Insights: This project leverages MRI imaging data to find
novel biomarkers for dementia classification.
• Innovation and Technology: The project employs advanced AI
techniques, including deep learning, to enhance the detection and
analysis of dementia stages.
3.
Objectives
• Accurate Classificationof Dementia Stages Using deep learning.
• Identification of Key Imaging Biomarkers.
• Enhancing Early Detection Capabilities.
4.
Background
• Context ofthe Project :The project uses deep learning
techniques to classify dementia into four categories—Mild
Demented, Moderate Demented, Very Mild Demented, and Non-
Demented—based on MRI images, supporting accurate and early
diagnosis.
• Significance: By applying state-of-the-art deep learning models,
this project enhances the precision of automated dementia
diagnosis, addressing the limitations of manual interpretation
while contributing to advancements in medical imaging and AI
integration in healthcare.
5.
Literature Survey
• Classificationof Alzheimer’s disease using MRI data based on Deep
Learning Techniques (2024)
[https://doi.org/10.1016/j.jksuci.2024.101940]
• Vision Transformer Approach for Classification of Alzheimer’s
Disease Using 18F-Florbetaben Brain Images (2023)
[https://doi.org/10.3390/app13063453]
• Classification of Alzheimer's Disease via Vision Transformer (2022)
[https://doi.org/10.1145/3529190.3534754]
• Ensemble of vision transformer architectures for efficient
Alzheimer’s Disease classification (2024)
[https://doi.org/10.1186/s40708-024-00238-7]
6.
Problem Identification
• Problemor Challenge Addressed
1. Difficulty in achieving early and accurate dementia classification due to
reliance on subjective clinical evaluations and manual MRI
interpretation.
2. Limited ability of existing methods to classify dementia into specific
subcategories, hindering tailored patient management.
• Limitations of Current Methods
1. Manual interpretation is time-consuming, prone to human error, and
unable to detect subtle early-stage patterns.
2. Existing technologies lack automation and scalability, making them
inefficient for large-scale deployment.
7.
Proposed Solution
• DeepLearning-Based Classification:
Utilizing a Convolutional Neural Network (CNN) architecture to analyze
MRI images and classify them into four categories: Mild Demented,
Moderate Demented, Very Mild Demented, and Non-Demented
• Concepts related to the chosen project.
Models that can be used
ResNet (Residual Networks)
VGGNet
DenseNet (Densely Connected Networks
U-Net
8.
Project Implementation Details
1.Data Collection and Preparation:
• Gather MRI images from verified sources with labels for Mild
Demented, Moderate Demented, Very Mild Demented, and Non-
Demented categories.
• Preprocess data by resizing images, normalizing pixel values, and
applying augmentation to handle class imbalances.
2. Model Development:
• Select a suitable deep learning architecture (e.g., CNN, ResNet, or
VGG).
• Implement transfer learning to leverage pre-trained models for
better feature extraction.
3. Training and Validation:
• Split the dataset into training, validation, and test sets.
• Perform hyperparameter tuning to optimize model performance.
9.
• Monitor metricslike accuracy, precision, recall, and F1-score
during training.
4. Model Evaluation:
• Test the trained model on unseen data to assess its generalization
capability.
• Use Grad-CAM or similar tools to validate predictions and provide
visual insights into classification decisions.
5. Deployment:
• Create a user-friendly interface for clinicians to upload MRI images
and receive automated classifications.
• Ensure scalability and robustness for real-world applications.