Scholarlink

Purpose <p>Early detection of cognitive decline is essential for timely diagnosis and treatment. This study aimed to evaluate whether upper-limb movement dynamics derived from wrist-mounted inertial sensors during a fine motor task can accurately differentiate among healthy aging, mild cognitive impairment (MCI), and dementia (DEM).</p> Methods <p>The study recruited young and older participants. Cognitive status was established through clinical evaluation together with concordance across three standardized neurocognitive assessments (MMSE, MoCA, and NEUROPSI). Upper-limb dynamics during the Grooved Pegboard Test (GPT) were recorded using a wrist-mounted triaxial BioHarness accelerometer. Time- and frequency-domain features were extracted from the accelerometry signals, and uncorrelated linear discriminant analysis (ULDA) was applied for supervised dimensionality reduction. Classification performance was evaluated using support vector machine (SVM), random forest (RF), and k-nearest neighbors (KNN) classifiers using accuracy and area under the receiver operating characteristic curve (AUC-ROC).</p> Results <p>The study included 109 participants distributed into four groups: 30 young healthy subjects (YNG), 22 cognitively intact older adults (CIO), 27 participants with MCI, and 30 with DEM. Prior to dimensionality reduction, the SVM classifier achieved the strongest performance, reaching a macro accuracy of 0.88 ± 0.013, demonstrating meaningful discrimination using the original feature representation. ULDA-based projection further improved class separability across all evaluated models. RF + ULDA and SVM + ULDA achieved macro accuracies of 0.97, whereas KNN + ULDA achieved 0.94.</p> Conclusion <p>Fine motor dynamics measured during the GPT contain discriminative information associated with progressive cognitive decline and may support objective, wearable-sensor-based cognitive screening.</p> Trial registration <p>The Local Ethics Committee of the General Hospital of Mexico approved the study, “Dr. Eduardo Liceaga” (protocol code DI/23/110-B/03/12).</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

Early Detection of Cognitive Impairment Using Time–Frequency Analysis of Fine Motor Accelerometry

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月22日

查看全文
  • Gustavo Pacheco-Santiago,
  • Itzel Iraís González-Aparicio,
  • Zeus Tlaltecutli Domínguez-Vega,
  • Lorena Velázquez-Álvarez,
  • Jonathan Roberto Torres-Castillo,
  • José de Jesús Rivera-Sánchez,
  • Miguel Ángel. Padilla-Castañeda

摘要

Purpose

Early detection of cognitive decline is essential for timely diagnosis and treatment. This study aimed to evaluate whether upper-limb movement dynamics derived from wrist-mounted inertial sensors during a fine motor task can accurately differentiate among healthy aging, mild cognitive impairment (MCI), and dementia (DEM).

Methods

The study recruited young and older participants. Cognitive status was established through clinical evaluation together with concordance across three standardized neurocognitive assessments (MMSE, MoCA, and NEUROPSI). Upper-limb dynamics during the Grooved Pegboard Test (GPT) were recorded using a wrist-mounted triaxial BioHarness accelerometer. Time- and frequency-domain features were extracted from the accelerometry signals, and uncorrelated linear discriminant analysis (ULDA) was applied for supervised dimensionality reduction. Classification performance was evaluated using support vector machine (SVM), random forest (RF), and k-nearest neighbors (KNN) classifiers using accuracy and area under the receiver operating characteristic curve (AUC-ROC).

Results

The study included 109 participants distributed into four groups: 30 young healthy subjects (YNG), 22 cognitively intact older adults (CIO), 27 participants with MCI, and 30 with DEM. Prior to dimensionality reduction, the SVM classifier achieved the strongest performance, reaching a macro accuracy of 0.88 ± 0.013, demonstrating meaningful discrimination using the original feature representation. ULDA-based projection further improved class separability across all evaluated models. RF + ULDA and SVM + ULDA achieved macro accuracies of 0.97, whereas KNN + ULDA achieved 0.94.

Conclusion

Fine motor dynamics measured during the GPT contain discriminative information associated with progressive cognitive decline and may support objective, wearable-sensor-based cognitive screening.

Trial registration

The Local Ethics Committee of the General Hospital of Mexico approved the study, “Dr. Eduardo Liceaga” (protocol code DI/23/110-B/03/12).