Better information, better decisions: Towards multi-modal integration of dementia biomarkers

First annual reportIn this initial stage, we performed a pilot analysis of the Dementia Disease Initiation (DDI) study. For that, we curated and processed resting-state Electroencephalogram (rsEEG), assessing the redundancy and relevance of different parameters extracted in this single modality prior to testing the workflow across different modalities.From September/2024 to January/2025 we transferred a total of 470 resting-state Electroencephalogram (rsEEG) recordings collected as part of the Dementia Disease Initiation study. This anonymized data was uploaded into a secure platform Tjenester for Sensitive Data (TSD) from the University of Oslo. Beyond data transferring, we have already preprocessed (curated) rsEEG from a total of (n = 160) individuals with available longitudinal or cross-sectional data. The extraction of different features from the rsEEG (slowing, complexity, connectivity) was also completed for this initial subsample. Subjects with established dementia-related neuropathology and clinical diagnosis were used to compare the rsEEG descriptors (grouping subjects with positive and negative amyloid, tau, neurodegeneration, apolipoprotein E4 alleles, and mild cognitive decline). The assessment of feature relevance and redundancy in the rsEEG was performed using the minimum-redundancy Maximum-Relevance algorithm implemented in the "featurewiz" library (https://github.com/AutoViML/featurewiz). The latter reduced significantly the number of rsEEG features, identifying meaningful features. We validated that the selected set of rsEEG features were comparable to those selected in an externally openly available large dataset of Alzheimer's Disease and preclinical stages (CAU-EEG Dataset, n = 965; https://www.sciencedirect.com/science/article/pii/S1053811923002008?via%3Dihub). As preliminary results, we have reduced the number of rsEEG features by 41,6%. This feature selection shows consistent results with group-related differences estimated via iterative inferential statistics. Finally, in subjects with positive mild cognitive impairment, amyloid, tau, or neurodegeneration in the DDI study, we observed that a combination of meaningful features comprising: slowing (in rhythmic and non-rhythmic activity) as well as complexity (greater signal regularity) in line with preliminary results presented in the PhD thesis that originated this postdoc project (https://uis.brage.unit.no/uis-xmlui/handle/11250/3145796)