Scholarlink

<p>Early symptoms of Alzheimer’s disease often manifest as Mild Cognitive Impairment making timely and crucial accurate classification for prompt diagnosis and therapeutic strategy development. However, mild cognitive impairment classification is hindered by challenges such as data heterogeneity, overfitting, and decreased learning efficiency, especially when handling complex, multimodal neuroimaging data. To tackle these issues, this research introduces a novel multimodal deep learning method for accurate mild cognitive impairment classification subtypes employing structural and functional MRI as well as demographic and cognitive features from Alzheimer’s disease Neuroimaging Initiative Dataset. This technique uses a graph convolutional network based multimodal hyper-connectivity network to capture complex intermodal relationships among extracted features. The structural MRI data undergo standardized processing including intensity normalization, skull stripping, spatial normalization and tissue segmentation while functional MRI data are processed into functional connectivity matrices and Pearson correlation. A hyper-connectivity graph is constructed employing mutual information and cross-correlation to quantify relationship between multimodal features. The graph convolutional network propagates and aggregates context-aware features which are further processed through customized deep neural network classifier employing ReLU activation, dropout and softmax output for multiclass classification. The simulation outcomes on the dataset demonstrate that the proposed multimodal deep learning technique significantly outperforms existing methods. It achieves accuracy (98.79%), precision (98.02%), recall (98.77%), specificity (98.76%), and F1-score (98.39%). These outcomes reflect the technique’s robust potential to support early detection and MCI monitoring which contributes to more efficient and accessible Alzheimer’s disease management.</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

Improving MCI classification with multimodal hyper-connectivity and deep neural networks for early detection of Alzheimer’s disease

  • 期刊论文

  • 发表日期：

    2025年10月18日

查看全文
  • Noorul Julaiha,
  • B. Vasudevan

摘要

Early symptoms of Alzheimer’s disease often manifest as Mild Cognitive Impairment making timely and crucial accurate classification for prompt diagnosis and therapeutic strategy development. However, mild cognitive impairment classification is hindered by challenges such as data heterogeneity, overfitting, and decreased learning efficiency, especially when handling complex, multimodal neuroimaging data. To tackle these issues, this research introduces a novel multimodal deep learning method for accurate mild cognitive impairment classification subtypes employing structural and functional MRI as well as demographic and cognitive features from Alzheimer’s disease Neuroimaging Initiative Dataset. This technique uses a graph convolutional network based multimodal hyper-connectivity network to capture complex intermodal relationships among extracted features. The structural MRI data undergo standardized processing including intensity normalization, skull stripping, spatial normalization and tissue segmentation while functional MRI data are processed into functional connectivity matrices and Pearson correlation. A hyper-connectivity graph is constructed employing mutual information and cross-correlation to quantify relationship between multimodal features. The graph convolutional network propagates and aggregates context-aware features which are further processed through customized deep neural network classifier employing ReLU activation, dropout and softmax output for multiclass classification. The simulation outcomes on the dataset demonstrate that the proposed multimodal deep learning technique significantly outperforms existing methods. It achieves accuracy (98.79%), precision (98.02%), recall (98.77%), specificity (98.76%), and F1-score (98.39%). These outcomes reflect the technique’s robust potential to support early detection and MCI monitoring which contributes to more efficient and accessible Alzheimer’s disease management.