Abstract:

Amnestic mild cognitive impairment (aMCI), a clinical prodromal stage, has a high probability of progressing to Alzheimer’s disease (AD). Therefore, new methods for the early identification and intervention are urgently required. Executive function deficits are characteristic of the initial stage of aMCI and may cause or aggravate memory symptoms, thereby increasing the risk of AD. Therefore, considering executive function as the entry point, it is possible to identify new targets for the early identification and intervention in aMCI. However, the neuropathological mechanisms underlying executive function in patients with aMCI remain unclear. The three subcomponents of inhibition, updating, and switching show both structural identity and specificity. To overcome the limitations of previous studies that viewed executive function as a whole or as separate elements, this study aims to focus on the overall structure of executive function and propose a hypothesis regarding the relationship between executive function and memory impairment in older patients with aMCI. It is also speculated that training the inhibitory subcomponents involving a larger range of frontoparietal control networks may be more helpful in attenuating or delaying memory impairment in patients with aMCI. Therefore, this study will follow the path of "assessment of spatiotemporal characteristics of cranial nerves - identification of characteristic targets - targeted digital intervention" for executive function in patients with aMCI, and will conduct verification by combining behavioral and ERP technology with deep learning and longitudinal intervention.
Study 1 will use high temporal resolution electroencephalography (EEG) technology as the main research method and adopt a two-factor mixed design; aMCI group vs. cognitively normal older individual group, and two stimulation types. By comparing the performance of Go/No-go, N-back, and set-switching tasks between the aMCI and normal cognitive older individual groups, the time-domain, time-frequency, and dynamic brain network characteristics of the patients with aMCI in the three executive function subcomponents of inhibition, updating, and switching will be systematically examined. Using this method, the neural mechanisms of these executive function subcomponents in patients with aMCI can be revealed, providing a new perspective for understanding their roles in memory impairment.
Study 2 will use deep learning to integrate the advantages of EEG temporal and spatial multidimensional data to construct a three-dimensional convolutional neural network classification model for the three executive function subcomponents of inhibition, updating, and switching in patients with aMCI. The focus is to examine the convolutional neural network classification model constructed based on the EEG characteristics of inhibitory function, which is a potential common cognitive process of different executive function subcomponents, namely the G factor. This study will analyze spatiotemporal features by extracting frequency-domain features, multi-layer convolution, and pooling layers from EEG data as inputs to the classifier. Through feature combination cluster analysis and 5-fold cross-validation, a model for aMCI inhibition, updating, and switching will be constructed, and cross-classification and cross-population validation conducted to evaluate the accuracy, sensitivity, and specificity of the model. It aims to more accurately identify and differentiate different executive function deficits in patients with aMCI and provide a new theoretical and practical basis for the early identification and treatment of aMCI.
To explore effective intervention strategies, this study also includes an analysis of the training effects and neural basis of different targeted digital interventions in patients with aMCI. Study 3 will adopt a 4 (inhibition group/refresh group/switching group/active control group) × 3 (measurement time: pretest/post-test 1/post-test 2) two-factor mixed design. The inhibition group, refresh group and switching group will use the difficulty-adaptive targeted digital interventions of "Whacking a Mole,” "Picturesque,” and "Dual Purpose,” respectively, while the active control group will only use processing speed training. Participants will be trained three times a week for 30 min each time, and post-tests will be conducted at weeks four and eight to examine intervention, transfer, and dose effects, as well as changes in corresponding neuroelectrophysiological indicators and dynamic brain network connections. The focus is on whether digital interventions targeting the inhibitory domain can improve episodic memory in patients with aMCI better than refreshing and switching targeted digital interventions.
This study aims to clarify that inhibition is a new cognitive target for executive function impairment and intervention in aMCI from a computational cognitive neural perspective, thereby providing an evidence-based basis for the early identification of aMCI and the formulation of precise diagnosis and treatment plans.

Key words: executive function, amnestic mild cognitive impairment, cognitive neural mechanism, digital intervention, deep learning