Scholarlink

Introduction: The research aims to construct artificial intelligence models for the classification of three states: Cognitive Normal (CN), Mild Cognitive Impairment (MCI), and Alzheimer’s Disease (AD), along with a binary classification of CN vs AD. This involves an investigation of various combinations of feature extraction, reduction, and classification techniques. The primary goal is to determine the most effective combination for cognitive state predictions. Methods: The study includes 601 MRI images (201 CN, 200 MCI, 201 AD) sourced from the Alzheimers Disease Neuroimaging Initiative (ADNI) dataset. The images are segmented into lateral ventricle and hippocampus regions. These segmented images are subjected to six types of 2D multiscale entropy. The remaining significant features are processed through Principal Component Analysis (PCA), Kernel Principal Component Analysis (KPCA), and Downsized Kernel Principal Component Analysis (DKPCA) for feature reduction. Finally, the features are input into Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), and K-Nearest Neighbor (KNN) and evaluated using precision, recall, and F1-score to find the best method. Results: For the CN vs MCI vs AD classification, the best combinations were KPCA-LDA and DKPCA-LDA, both achieving an F1-score of 0.62. CN vs AD exhibited higher performance across all combinations, with DKPCA-LDA achieving the highest F1-score of 0.87. The confusion matrix of the highest-performing model, LDA, indicates that the MCI was most frequently misclassified, while CN versus AD. The classification performance showed lower results when using only hippocampus features compared to using lateral ventricle features (0.55 versus 0.62 for three classes, 0.80 versus 0.83 for two classes). Conclusion and Discussion: The study emphasizes the potential of DKPCA-LDA for multistage Alzheimer’s classification using MRI images, achieving the highest classification accuracy. However, the lower performance of the hippocampus compared to the lateral ventricle underscores the challenges of Alzheimer’s classification. While the hippocampus is a logical feature from a neurological standpoint, its similar results across different Alzheimer’s stages make it less rational from a computer vision perspective.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multi-stage Alzheimer’s Diagnosis Using MRI Image: A Combination of Computer Vision Technique for Machine Learning Model

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Tri Minh Huynh Nguyen,
  • Tra My Nguyen Thi,
  • Huong Thi Thanh Ha,
  • Nhat Thanh Nguyen Truong,
  • Tuong Huu Nguyen,
  • Chien Phan Cong,
  • Thu Minh Phan Ngoc

摘要

Introduction: The research aims to construct artificial intelligence models for the classification of three states: Cognitive Normal (CN), Mild Cognitive Impairment (MCI), and Alzheimer’s Disease (AD), along with a binary classification of CN vs AD. This involves an investigation of various combinations of feature extraction, reduction, and classification techniques. The primary goal is to determine the most effective combination for cognitive state predictions. Methods: The study includes 601 MRI images (201 CN, 200 MCI, 201 AD) sourced from the Alzheimers Disease Neuroimaging Initiative (ADNI) dataset. The images are segmented into lateral ventricle and hippocampus regions. These segmented images are subjected to six types of 2D multiscale entropy. The remaining significant features are processed through Principal Component Analysis (PCA), Kernel Principal Component Analysis (KPCA), and Downsized Kernel Principal Component Analysis (DKPCA) for feature reduction. Finally, the features are input into Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), and K-Nearest Neighbor (KNN) and evaluated using precision, recall, and F1-score to find the best method. Results: For the CN vs MCI vs AD classification, the best combinations were KPCA-LDA and DKPCA-LDA, both achieving an F1-score of 0.62. CN vs AD exhibited higher performance across all combinations, with DKPCA-LDA achieving the highest F1-score of 0.87. The confusion matrix of the highest-performing model, LDA, indicates that the MCI was most frequently misclassified, while CN versus AD. The classification performance showed lower results when using only hippocampus features compared to using lateral ventricle features (0.55 versus 0.62 for three classes, 0.80 versus 0.83 for two classes). Conclusion and Discussion: The study emphasizes the potential of DKPCA-LDA for multistage Alzheimer’s classification using MRI images, achieving the highest classification accuracy. However, the lower performance of the hippocampus compared to the lateral ventricle underscores the challenges of Alzheimer’s classification. While the hippocampus is a logical feature from a neurological standpoint, its similar results across different Alzheimer’s stages make it less rational from a computer vision perspective.