Scholarlink

<p>Alzheimer’s Disease (AD) is a progressive, complex, and multifaceted neurological disorder that affects the brain. Identification of Alzheimer’s disease at an early stage is the most effective way of preventing and controlling its progression. Therefore, intelligent automated systems (especially those based on machine learning) have become a prominent and indispensable tool. The present study aims to develop an automatic system for the classification of AD through the use of manual feature extraction, hyperparameter tuning, and probability fusion to strengthen the diagnostic accuracy and speed. The research utilizes a publicly available dataset from the Alzheimer’s Disease Neuroimaging Initiative, which includes MRI images that have been spatially normalized, masked, and N3-corrected, as well as T1-weighted images. Local and global brain changes were captured by the hand-crafted features that were extracted. To ensure stability of the predictions, a fusion layer was created to merge the outputs of the individual models in terms of probability. The experimental results demonstrate that the fusion approach outperforms each feature category. Precision and specificity were both at 100% (perfect) with fusion, while F1-score, accuracy, and average scores were also very close to 100% (96.44% and 97.24%, respectively). Furthermore, GLCM and GLRLM were the best-performing individual feature categories, as measured by average scores of 81% and 79.5%, respectively. The statistical test supported the stability of the fusion model, as indicated by the very low variations in performance metrics and the narrow confidence intervals.</p>

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

Evaluating the role of 3D hand-crafted fused features in Alzheimer’s disease classification from a machine learning perspective

  • 期刊论文

  • 发表日期：

    2026年3月14日

查看全文
  • Saleh Ateeq Almutairi

摘要

Alzheimer’s Disease (AD) is a progressive, complex, and multifaceted neurological disorder that affects the brain. Identification of Alzheimer’s disease at an early stage is the most effective way of preventing and controlling its progression. Therefore, intelligent automated systems (especially those based on machine learning) have become a prominent and indispensable tool. The present study aims to develop an automatic system for the classification of AD through the use of manual feature extraction, hyperparameter tuning, and probability fusion to strengthen the diagnostic accuracy and speed. The research utilizes a publicly available dataset from the Alzheimer’s Disease Neuroimaging Initiative, which includes MRI images that have been spatially normalized, masked, and N3-corrected, as well as T1-weighted images. Local and global brain changes were captured by the hand-crafted features that were extracted. To ensure stability of the predictions, a fusion layer was created to merge the outputs of the individual models in terms of probability. The experimental results demonstrate that the fusion approach outperforms each feature category. Precision and specificity were both at 100% (perfect) with fusion, while F1-score, accuracy, and average scores were also very close to 100% (96.44% and 97.24%, respectively). Furthermore, GLCM and GLRLM were the best-performing individual feature categories, as measured by average scores of 81% and 79.5%, respectively. The statistical test supported the stability of the fusion model, as indicated by the very low variations in performance metrics and the narrow confidence intervals.