Scholarlink

<p>Alzheimer’s disease (AD) and frontotemporal dementia (FTD) are two major causes of dementia, with overlapping clinical and pathophysiological characteristics. We investigated whether resting-state EEG functional connectivity could distinguish AD, FTD, and healthy controls (HC) using a stacked ensemble learning approach. A publicly available dataset (openneuro.org/datasets/ds004504) was used to extract multiple connectivity metrics across frequency bands. Base classifiers were trained using the Fisher’s Geodesic Minimum Distance to Mean (FgMDM) approach with Euclidean or Riemannian distance metrics, and a meta-classifier was trained on their cross-validated predictions. Using leave-one-subject-out cross-validation, the stacked model achieved ROC-AUC values of 81.80% and 71.36% for distinguishing AD and FTD from HC, respectively. This was lower than the best single-base classifier performance, where several connectivity metrics achieved ROC-AUC values of 85% or higher for distinguishing AD and FTD. AD-FTD discrimination proved most challenging (ROC-AUC of 65.10%), suggesting overlapping network disruptions that limit separability. Classification performance was highest in the alpha band when distinguishing AD and FTD from HC, whereas the highest performance for distinguishing AD from FTD was observed in the delta band. Overall, these findings highlight the potential of EEG connectivity features for dementia classification and emphasize the importance of careful feature engineering and low-complexity models in the case of small-sample datasets.</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

EEG-based classification of alzheimer’s disease and frontotemporal dementia using functional connectivity

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月9日

查看全文
  • Tjaša Mlinarič,
  • Arne Van Den Kerchove,
  • Zoe I. Barinaga,
  • Marc M. Van Hulle

摘要

Alzheimer’s disease (AD) and frontotemporal dementia (FTD) are two major causes of dementia, with overlapping clinical and pathophysiological characteristics. We investigated whether resting-state EEG functional connectivity could distinguish AD, FTD, and healthy controls (HC) using a stacked ensemble learning approach. A publicly available dataset (openneuro.org/datasets/ds004504) was used to extract multiple connectivity metrics across frequency bands. Base classifiers were trained using the Fisher’s Geodesic Minimum Distance to Mean (FgMDM) approach with Euclidean or Riemannian distance metrics, and a meta-classifier was trained on their cross-validated predictions. Using leave-one-subject-out cross-validation, the stacked model achieved ROC-AUC values of 81.80% and 71.36% for distinguishing AD and FTD from HC, respectively. This was lower than the best single-base classifier performance, where several connectivity metrics achieved ROC-AUC values of 85% or higher for distinguishing AD and FTD. AD-FTD discrimination proved most challenging (ROC-AUC of 65.10%), suggesting overlapping network disruptions that limit separability. Classification performance was highest in the alpha band when distinguishing AD and FTD from HC, whereas the highest performance for distinguishing AD from FTD was observed in the delta band. Overall, these findings highlight the potential of EEG connectivity features for dementia classification and emphasize the importance of careful feature engineering and low-complexity models in the case of small-sample datasets.