The Use of Speech in Screening for Cognitive Decline in Older Adults
10.16476/j.pibb.2024.0329
- VernacularTitle:言语在筛查老年人认知功能下降中的应用
- Author:
Si-Wen WANG
1
;
Xiao-Xiao YIN
1
;
Lin-Lin GAO
2
;
Wen-Jun GUI
1
;
Qiao-Xia HU
3
;
Qiong LOU
3
;
Qin-Wen WANG
4
Author Information
1. School of Teacher Education, Ningbo University, Ningbo 315211, China
2. College of Information Science and Engineering, Ningbo University, Ningbo 315211, China
3. The First Affiliated Hospital of Ningbo University, Ningbo 315211, China
4. School of Basic Medicine, School of Medicine, Ningbo University, Ningbo 315211, China
- Publication Type:Journal Article
- Keywords:
Alzheimer’s disease;
cognitive function decline;
early screening;
speech task
- From:
Progress in Biochemistry and Biophysics
2025;52(2):456-463
- CountryChina
- Language:Chinese
-
Abstract:
Alzheimer’s disease (AD) is a chronic neurodegenerative disorder that severely affects the health of the elderly, marked by its incurability, high prevalence, and extended latency period. The current approach to AD prevention and treatment emphasizes early detection and intervention, particularly during the pre-AD stage of mild cognitive impairment (MCI), which provides an optimal “window of opportunity” for intervention. Clinical detection methods for MCI, such as cerebrospinal fluid monitoring, genetic testing, and imaging diagnostics, are invasive and costly, limiting their broad clinical application. Speech, as a vital cognitive output, offers a new perspective and tool for computer-assisted analysis and screening of cognitive decline. This is because elderly individuals with cognitive decline exhibit distinct characteristics in semantic and audio information, such as reduced lexical richness, decreased speech coherence and conciseness, and declines in speech rate, voice rhythm, and hesitation rates. The objective presence of these semantic and audio characteristics lays the groundwork for computer-based screening of cognitive decline. Speech information is primarily sourced from databases or collected through tasks involving spontaneous speech, semantic fluency, and reading, followed by analysis using computer models. Spontaneous language tasks include dialogues/interviews, event descriptions, narrative recall, and picture descriptions. Semantic fluency tasks assess controlled retrieval of vocabulary items, requiring participants to extract information at the word level during lexical search. Reading tasks involve participants reading a passage aloud. Summarizing past research, the speech characteristics of the elderly can be divided into two major categories: semantic information and audio information. Semantic information focuses on the meaning of speech across different tasks, highlighting differences in vocabulary and text content in cognitive impairment. Overall, discourse pragmatic disorders in AD can be studied along three dimensions: cohesion, coherence, and conciseness. Cohesion mainly examines the use of vocabulary by participants, with a reduction in the use of nouns, pronouns, verbs, and adjectives in AD patients. Coherence assesses the ability of participants to maintain topics, with a decrease in the number of subordinate clauses in AD patients. Conciseness evaluates the information density of participants, with AD patients producing shorter texts with less information compared to normal elderly individuals. Audio information focuses on acoustic features that are difficult for the human ear to detect. There is a significant degradation in temporal parameters in the later stages of cognitive impairment; AD patients require more time to read the same paragraph, have longer vocalization times, and produce more pauses or silent parts in their spontaneous speech signals compared to normal individuals. Researchers have extracted audio and speech features, developing independent systems for each set of features, achieving an accuracy rate of 82% for both, which increases to 86% when both types of features are combined, demonstrating the advantage of integrating audio and speech information. Currently, deep learning and machine learning are the main methods used for information analysis. The overall diagnostic accuracy rate for AD exceeds 80%, and the diagnostic accuracy rate for MCI also exceeds 80%, indicating significant potential. Deep learning techniques require substantial data support, necessitating future expansion of database scale and continuous algorithm upgrades to transition from laboratory research to practical product implementation.
