Frontiers in Aging Neuroscience · Nov 16, 2026

Acoustic Features for Assisting Mild Cognitive Impairment Detection: A Cross-Sectional Diagnostic Accuracy Study

By Yue Liu, Mingjing Yang, Qiaoyang Zhang et al.

Background: Mild cognitive impairment (MCI) is a key preclinical stage of Alzheimer's disease (AD). Current scale-based assessments are subjective, expertise-dependent, and poorly suited for primary care. Acoustic features may offer objective markers of early decline. Objective: This study aims to develop a non-invasive, auxiliary model for recognizing mild cognitive impairment (MCI), which is a critical precursor to Alzheimer's disease, using the "Cookie Theft" picture description task. Methods: This study included 703 participants, with 497 recruited from Changzhou and 206 from Nanjing. Speech samples were collected using the “Cookie Theft”task. An 88-dimensional eGeMAPS acoustic feature set was extracted. A demographic model and a demographic-acoustic fusion model were developed using nine machine learning algorithms. Model performance was evaluated primarily using the area under the curve (AUC), complemented by sensitivity and specificity. External generalizability was also assessed. Results: The LR-based fusion model demonstrated robust and balanced performance, yielding an internal test AUC of 0.820 (0.741-0.889), sensitivity of 0.703 (SD = 0.052), and specificity of 0.745 (SD = 0.063). External independent validation yielded an AUC of 0.679 (0.600–0.756), with sensitivity of 0.678 (SD = 0.038) and specificity of 0.600 (SD = 0.064). Conclusion: Acoustic features enhance the identification performance of MCI, providing preliminary support for a low-cost, scalable, non-invasive screening approach for MCI auxiliary recognition in primary care settings.

The paperJournal siteBiomarkers

AgingResearch.News