Deep learning for diagnosis of mild cognitive impairment in older adults:a scoping review
10.3969/j.issn.1006-9771.2025.06.008
- VernacularTitle:深度学习在老年人轻度认知障碍诊断中应用的Scoping综述
- Author:
Xiaohui WU
1
;
Lei JIANG
;
Jingru ZHU
;
Lihui XIE
Author Information
1. 首都体育学院实验室与国有资产管理处,北京市 100191
- Publication Type:Journal Article
- Keywords:
older adults;
mild cognitive impairment;
deep learning;
scoping review
- From:
Chinese Journal of Rehabilitation Theory and Practice
2025;31(6):674-681
- CountryChina
- Language:Chinese
-
Abstract:
Objective To systematically review the application and effectiveness of deep learning(DL)in diagnosis of mild cogni-tive impairment(MCI)among older adults.Methods PubMed,Web of Science,CNKI and Wanfang databases were searched for literatures related to the application of DL in MCI among older adults,from database inception to December,2024.A scoping review was conducted.The literature screening process followed the Scoping Review Report Specification list,and the quality assess-ment was conducted using the cross-sectional study quality evaluation tool developed by the Evidence-based Health Care Center.Results A total of eleven papers were included,from Italy,USA,South Korea,China,India and Switzerland,involving 11 829 elderly participants,publicated mainly between 2014 and 2024,reflecting the rapid development trend of the field in the last decade,which was in line with the timing of the development of DL technology.The quality scores of the included literatures were all six to seven.The types of studies were all cross-sectional studies with significant cross-disciplinary characteristics,mainly originating from the fields of clinical medicine,biology and neuroimaging.The literature data were mainly based on the Alzheimer's disease Neuroimaging Program database and integrated other data resources.In terms of data type,in addition to brain imaging data,one study based on text data was also included.In terms of models used,five of the studies were mainly based on convolutional neu-ral networks,and the rest used different DL modeling frameworks.The task types contained binary and triple classification.In terms of prediction results,the DL models constructed on multimodal data,such as brain imag-es,could be used to construct high-precision prediction models for MCI classification,and the models were all good,with accuracy more than 70%and AUC values more than 0.7.The diagnostic accuracy of some of the mod-els was more than 90%,and the model with the highest prediction accuracy was the one that used the Biceph-Net lightweight framework,with accuracy close to 100%,and the text analysis model based on Transformer made the AUC value of 0.846,which provided new ideas for the diagnosis of non-imaging data.Conclusion DL can not only provide strong support for the accurate identification of MCI in the elderly,but also provide auxiliary prediction tools for clinicians,which can help delay the progression of the disease and improve the prognosis of patients.
