IMPROVING ALZHEIMER''S DISEASE DIAGNOSIS WITH CONVOLUTIONAL NEURAL NETWORKS LEVERAGING ADVANCED MAXPO

Abstract

Alzheimer’s disease (AD) is a major public health priority. Hippocampus is one of the most affected areas of the brain and is easily accessible as a biomarker using MRI images in machine learning for diagnosing AD. In machine learning, using entire MRI image slices showed lower accuracy for AD classification. We present the select slices method by landmarks on the hippocampus region in MRI images. This study aims to see which views of MRI images have higher accuracy for AD classification. Then, to get the value of three views and categories, we used multiclass classification with the publicly available Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset using Resnet50 and LeNet. The models were used in a total dataset of 4,500 MRI slices in three views and categories. Our study demonstrated that the selecting slices performed better than using entire slices in MRI images for AD classification. Our method improves the accuracy of machine learning, and the coronal view showed higher accuracy. This method played a significant role in improving the accuracy of machine learning performance. The results for the coronal view were similar to the medical experts usually used to diagnose AD. We also found that LeNet models became the potential model for AD classification.

Specification

Field o f Invention: Convolutional Neural Networks
The invention lies at the intersection of artificial intelligence, medical imaging, and neurodiagnostics, focusing on improving Alzheimer's Disease diagnosis. By applying advanced Convolutional Neural Networks (CNNs) to hippocampal MRI scans, it leverages enhanced MaxPooling and Dropout techniques to extract critical features, reduce overfitting, and increase diagnostic accuracy. This innovation integrates deep, learning with medical imaging to provide a more reliable and efficient early detection tool for Alzheimer's, aiming to advance Al-driven diagnostics in healthcare.
Background Art including citations o f prior art: Prior research has shown that using whole-brain MRI scans for Alzheimer's Disease (AD) diagnosis with machine learning often reduces accuracy due to irrelevant features outside key brain regions. Studies like Liu et al. (2018) and Cheng et al. (2020) emphasized the value of focusing on the hippocampus, a crucial area affected early in AD. However, they did not fully explore the impact of different MRI views. Our approach builds on this by selecting specific slices of the hippocampus in Axial, Coronal, and Sagittal views, demonstrating that targeted views-particularly the coronal-improve classification accuracy using models like ResnetSO and LeNet, consistent with findings by Zhou et al. (2022) on multi-view analysis. This method refines diagnostic precision, aligning closely with expert assessments.
1
Objective o f invention (the invention's objectives and advantages, or alternative
embodiments of the invention):
The primary objective of this invention is to improve the accuracy of Alzheimer's Disease (AD) diagnosis by utilizing a novel approach that selectively analyzes specific MRI slices of the

hippocampus in multiple views (Axial, Coronal, and Sagittal). This method aims to enhance classification performance in machine learning models, particularly by employing advanced deep learning architectures like Resnet50 and LeNet.
Key Objectives:
1. Enhance Diagnostic Accuracy: To achieve higher accuracy in AD classification by focusing on the hippocampus, which is significantly affected by the disease. By analyzing selected MRJ slices rather than entire scans, the invention minimizes the influence of irrelevant features, thus improving model performance. 2. Evaluate MRI View Impact: To determine which specific views of MRJ images (Axial, Coronal, and Sagittal) provide the best classification results for AD, with a focus on identifying the coronal view as the most effective for this purpose. 3. Optimize Machine Learning Models: To compare the performance of different deep learning algorithms (including Pre-trained Resnet50, ResnetSO, and LeNet) to establish the most effective architecture for AD classification, demonstrating that LeNet achieves
superior accuracy.
Advantages:
• Improved Precision: By selecting targeted slices, the proposed method significantly enhances diagnostic accuracy compared to existing systems that use whole MRI scans. • Alignment with Expert Analysis: The coronal view's accuracy mirrors that of medical experts, supporting the clinical relevance of the proposed technique. • Efficient Data Utilization: The approach reduces data complexity by eliminating irrelevant features, leading to more efficient training and faster processing times for
machine learning models.
Alternative Embodiments:
1. Integration with Other Biomarkers: The invention could be adapted to incorporate additional biomarkers from other regions of the brain or other imaging modalities (e.g.,11+Nov-2024/135134/202441086596/Form 2(Title Page)
2. Expansion to Other Neurodegenerative Diseases: The methodology could be modified to apply to the diagnosis of other neurodegenerative disorders where specific brain regions are affected, thus broadening the impact of the invention. 3. Real-Time Diagnostic Tools: Developing a software tool for real-time analysis of MRI scans using the proposed method, enabling clinicians to leverage machine learning
insights during patient assessments.
4. User-Friendly Interfaces: Creating intuitive interfaces for radiologists and clinicians to easily interpret the model's findings and integrate them into clinical workflows.
These objectives and advantages position this invention as a significant advancement in the field of Alzheimer's Disease diagnosis, promising better outcomes through targeted analysis and cutting-edge machine learning techniques.
Summary of Invention:
The invention proposes a novel method for enhancing the diagnosis of Alzheimer's Disease (AD) through the selective analysis of MRI images, specifically targeting the hippocampal region. Recognizing that traditional approaches utilizing entire MRI scans often lead to reduced classification accuracy due to the inclusion of irrelevant features, this invention focuses on extracting and evaluating MR] slices from three specific views: Axial, Coronal, and Sagittal.
By employing advanced deep learning models, including Pre-trained Resnet50, ResnetSO, and LeNet, the invention systematically analyzes the impact of different MRI views on diagnostic accuracy. The study demonstrates that targeted slice selection significantly improves machine learning performance, with the coronal view yielding the highest accuracy comparable to expert
assessments.
The advantages of this method include enhanced precision in AD classification, efficient data utilization by reducing unnecessary complexity, and alignment with clinical practices.
Additionally, the findings indicate that LeNet emerges as a particularly effective model for this


This invention not only contributes to the field of neuroimaging and machine learning but also holds promise for further applications in diagnosing other neurodegenerative diseases. By providing a robust, accurate, and clinically relevant diagnostic tool, this invention aims to improve early detection and intervention strategies for Alzheimer's Disease, ultimately
benefiting patient care and outcomes.
Detailed description of the invention:
The present invention focuses on enhancing the diagnosis of Alzheimer's Disease (AD) through a novel approach that utilizes specific MRI slices from the hippocampus region of the brain. This method leverages advanced deep learning techniques to optimize diagnostic accuracy and improve the overall efficacy of machine learning models in identifying AD.
1. Background and Motivation
Alzheimer's Disease is a major public health concern, with increasing prevalence and a significant impact on cognitive health. Early and accurate diagnosis is crucial for effective management and treatment. Traditional diagnostic methods often rely on cognitive assessments and neuroimaging, but using whole MRI scans can introduce noise and irrelevant data, which can degrade the classification performance of machine learning models. 2. Objectives of the Invention
The invention aims to address these challenges by focusing on the following objectives: • Selective Slice Analysis: To improve classification accuracy by analyzing targeted MRI slices from the hippocampus in specific views (Axial, Coronal, and Sagittal). • Performance Comparison: To evaluate the effectiveness of different deep learning models (Pre-trained Resnet50, Resnet50, and LeNet) on the selected slices. • Identification of Optimal Views: To determine which MRI view yields the highest accuracy for AD classification, particularly highlighting the coronal view.
3. Methodology
Neuroimaging Initiative (ADNI) dataset, comprising a total of 4.500 MRI slices across the specified views and categories. This dataset provides a diverse representation of various stages of Alzheimer's, facilitating robust model training and evaluation. • Slice Selection: MRJ images are pre-processed to focus on the hippocampal region, where the most significant changes associated with AD occur. Using landmark-based techniques, the invention identifies and extracts relevant slices from each MRI scan, ensuring that the analysis is concentrated on areas most indicative of disease
progression.
• Deep Learning Models: The invention employs three distinct deep learning
architectures:
o Pre-trained Resnet50: A deep residual network that leverages transfer learning to utilize knowledge from previously trained models, enhancing performance on
smaller datasets.
o ResnetSO: A standard implementation of the Resnet architecture that utilizes skip connections to facilitate training deeper networks without suffering from
vanishing gradients.
o LeNet: A simpler convolutional neural network model known for its effectiveness in image classification tasks, particularly suitable for smaller
datasets.
• Training and Evaluation: Each model is trained using the selected MRI slices, and performance metrics such as accuracy, sensitivity, and specificity are computed to evaluate the effectiveness of each architecture. The invention highlights that the LeNet model achieves the best results, demonstrating its potential for AD classification.
4. Results and Findings
The experimental results demonstrate that the selective slice method outperforms existing systems that utilize entire MRI scans. Key findings include: • Improved Accuracy: The selective analysis,, particularly ip .the, coronal view, shows


significant increase in classification accuracy compared to using whole MRI slices. • Alignment with Expert Diagnosis: The classification results align closely with those of medical experts, validating the clinical relevance of the approach. • Model Performance: Among the models tested, LeNet shows exceptional promise, providing a balance of accuracy and computational efficiency.
5. Advantages of the Invention
• . Enhanced Diagnostic Precision: By focusing on relevant features within targeted slices, the invention improves the overall accuracy of Alzheimer's diagnosis. • Efficient Data Handling: The method reduces the complexity of MRI data, allowing for more efficient training of machine learning models. • Clinical Applicability: The findings support the integration of this method into clinical workflows, offering a reliable tool for early diagnosis and intervention in Alzheimer's
Disease.
6. Potential Applications and Future Work The invention not only addresses current limitations in AD diagnosis but also opens avenues for
further research, including:
• Application to Other Neurodegenerative Diseases: The methodology could be adapted for the diagnosis of other conditions where specific brain regions are critical. • Real-Time Diagnostic Tools: Developing software for real-time analysis of MRI scans using the proposed techniques could enhance clinical decision-making. • Extended Model Testing: Future work may involve testing additional deep learning architectures and techniques, such as ensemble methods or attention mechanisms, to
further optimize performance.

Claims:
1. A method for improving Alzheimer's classification using MRI, involving selecting specific hippocampal slices and applying deep learning algorithms. 2. The method of claim 1, wherein MRI views are divided into axial, coronal, and sagittal
for targeted analysis.
3. The method of claim 1, wherein the coronal view is prioritized due to its superior classification accuracy for Alzheimer's. 4. A system for implementing the method of claim 1, using a multiclass classification framework with ResNetSO and LeNet models. 5. The system of claim 4, wherein LeNet shows better accuracy when analyzing selected
hippocampal slices. .
6. The system of claim 4, utilizing a dataset of 4,500 MRI slices from ADNI to validate
accuracy.
- -7~The method of' clainTl. minimizing irrelevant features in MRI images to reduce. false positives and enhance machine learning predictions for Alzheimer's.

Documents