Scholarlink

<p>Cognitive reserve mitigates aging-related cognitive decline and frailty, yet current assessments lack neurobiological specificity. We aimed to develop a noninvasive, functional near infrared spectroscopy (fNIRS)–based machine learning model to classify cognitive reserve levels in older adults with cognitive frailty. Seventy-one community-dwelling adults underwent resting-state and task-based (Stroop, <i>n</i>-back) fNIRS scans. Graph theory metrics and task-related <i>β</i>-values were extracted. Support vector machine classifiers were trained on 70% of the dataset and tested on 30%. Models incorporating <i>β</i>-values from significantly activated channels during the Stroop, 0-back, and 1-back tasks achieved the best performance (accuracy = 0.727, recall = 0.857, area under the curve [AUC] = 0.829). Resting-state features alone yielded lower performance (AUC = 0.714), while combining both resting-state and task-based features improved it moderately (AUC = 0.790). fNIRS-based modeling enables objective classification of cognitive reserve levels in older adults with cognitive frailty. This approach offers a portable, scalable, real-time strategy for early risk stratification and may support precision interventions in both clinical and community settings.</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

Machine learning assessment of cognitive reserve using functional near-infrared spectroscopy in older adults with cognitive frailty

  • 期刊论文

  • 发表日期：

    2025年10月12日

查看全文
  • Wanrui Wei,
  • Shuaifang Wei,
  • Wei Han,
  • Kairong Wang,
  • Huan Zhang,
  • Gabriella Engstrom,
  • Azita Emami,
  • Zheng Li

摘要

Cognitive reserve mitigates aging-related cognitive decline and frailty, yet current assessments lack neurobiological specificity. We aimed to develop a noninvasive, functional near infrared spectroscopy (fNIRS)–based machine learning model to classify cognitive reserve levels in older adults with cognitive frailty. Seventy-one community-dwelling adults underwent resting-state and task-based (Stroop, n-back) fNIRS scans. Graph theory metrics and task-related β-values were extracted. Support vector machine classifiers were trained on 70% of the dataset and tested on 30%. Models incorporating β-values from significantly activated channels during the Stroop, 0-back, and 1-back tasks achieved the best performance (accuracy = 0.727, recall = 0.857, area under the curve [AUC] = 0.829). Resting-state features alone yielded lower performance (AUC = 0.714), while combining both resting-state and task-based features improved it moderately (AUC = 0.790). fNIRS-based modeling enables objective classification of cognitive reserve levels in older adults with cognitive frailty. This approach offers a portable, scalable, real-time strategy for early risk stratification and may support precision interventions in both clinical and community settings.