Scholarlink

<p>Alzheimer’s disease (AD) is a degenerative ailment that causes the brain to shrink and ultimately experience cell death. AD affects not only the individual but also families, caregivers, and society. The diagnosis of AD at early stages is challenging but possible with the help of multimodality imaging, such as magnetic resonance imaging (MRI) and fluorodeoxyglucose (FDG)-positron emission tomography (PET). In this study, a lightweight, custom-designed convolutional neural network (CNN) architecture tailored to classify four distinct stages of AD (NonDemented, VeryMildDemented, MildDemented, and ModerateDemented) using MRI scans is proposed. Unlike existing approaches, the dataset was enhanced by integrating under-represented classes from multiple public sources (Kaggle and ADNI), achieving a balanced dataset of 10,114 images. The baseline model is systematically evaluated against six well-known pre-trained architectures (VGG16, VGG19, ResNet50, ResNet101, Xception, and InceptionV3), demonstrating superior or comparable performance, especially in early-stage classification. The baseline model achieved an accuracy of 97.80% using 5-fold cross-validation, with high consistency across SMOTE-augmented and non-augmented cases. To ensure clinical relevance, Grad-CAM approach is employed to generate plausible regions of interest with heatmaps in the brain. These findings underscore the effectiveness of a custom-designed CNN for accurate staging of AD, offering potential clinical utility in supporting diagnostic decisions, therapeutic planning, and personalized care strategies. The proposed model and its implementation are publicly available via GitHub, offering a transparent, interpretable, and practical tool for AD stage diagnosis.</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

Explainable AI-driven Alzheimer’s disease stage diagnosis leveraging CNN

  • 期刊论文

  • 发表日期：

    2026年3月14日

查看全文
  • Dhairiya Agarwal,
  • Anju Sharma,
  • Prabha Garg

摘要

Alzheimer’s disease (AD) is a degenerative ailment that causes the brain to shrink and ultimately experience cell death. AD affects not only the individual but also families, caregivers, and society. The diagnosis of AD at early stages is challenging but possible with the help of multimodality imaging, such as magnetic resonance imaging (MRI) and fluorodeoxyglucose (FDG)-positron emission tomography (PET). In this study, a lightweight, custom-designed convolutional neural network (CNN) architecture tailored to classify four distinct stages of AD (NonDemented, VeryMildDemented, MildDemented, and ModerateDemented) using MRI scans is proposed. Unlike existing approaches, the dataset was enhanced by integrating under-represented classes from multiple public sources (Kaggle and ADNI), achieving a balanced dataset of 10,114 images. The baseline model is systematically evaluated against six well-known pre-trained architectures (VGG16, VGG19, ResNet50, ResNet101, Xception, and InceptionV3), demonstrating superior or comparable performance, especially in early-stage classification. The baseline model achieved an accuracy of 97.80% using 5-fold cross-validation, with high consistency across SMOTE-augmented and non-augmented cases. To ensure clinical relevance, Grad-CAM approach is employed to generate plausible regions of interest with heatmaps in the brain. These findings underscore the effectiveness of a custom-designed CNN for accurate staging of AD, offering potential clinical utility in supporting diagnostic decisions, therapeutic planning, and personalized care strategies. The proposed model and its implementation are publicly available via GitHub, offering a transparent, interpretable, and practical tool for AD stage diagnosis.