Scholarlink

Alzheimer’s disease is a well-known neurological condition that is the most common cause of dementia. It affects older adults leading to a decline in cognitive function, and memory which affects daily activities. This study investigates which combination between encoding technique (One-Hot, Label, Binary, Ordinal, and Leave-one-out) for the dataset and activation functions (ReLU, LeakyReLU) in the CNN model delivers the best performance, when Precision, Recall and F1score are evaluated. The different CNN models were trained over 10, 50 and 100 epochs. By using a publicly available MRI scans dataset, with four categories of Alzheimer’s disease, Non_Demented, Mild_Demented, Moderate_Demented, and Very_Mild_Demented. The study shows that an accuracy of 100% can be obtained when OneHot and LeakyReLU are employed. The results outperform similar research, where the same dataset, encoding techniques and activation functions are used.

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

Accurate MRI-Based Alzheimer’s Disease Detection Using CNNs and Encoding Techniques

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Florina-Luminița Rostaș,
  • Raluca-Cosmina-Maria Ardelean,
  • Laura-Nicoleta Ivanciu,
  • Emilia Șipoș

摘要

Alzheimer’s disease is a well-known neurological condition that is the most common cause of dementia. It affects older adults leading to a decline in cognitive function, and memory which affects daily activities. This study investigates which combination between encoding technique (One-Hot, Label, Binary, Ordinal, and Leave-one-out) for the dataset and activation functions (ReLU, LeakyReLU) in the CNN model delivers the best performance, when Precision, Recall and F1score are evaluated. The different CNN models were trained over 10, 50 and 100 epochs. By using a publicly available MRI scans dataset, with four categories of Alzheimer’s disease, Non_Demented, Mild_Demented, Moderate_Demented, and Very_Mild_Demented. The study shows that an accuracy of 100% can be obtained when OneHot and LeakyReLU are employed. The results outperform similar research, where the same dataset, encoding techniques and activation functions are used.