Aims

The aim of the study is to verify whether EEG biomarkers can be used to classify patients with dementia from controls and patients with AD dementia versus patients with non-AD dementia.

Methods

EEG measurements of 53 AD patients, 49 patients with a different type of non-AD dementia and 83 controls are used. After preprocessing, multiple EEG features, such as spectral power and functional brain connectivity are calculated. Support vector machines (SVM) are applied to classify (i) dementia vs. controls and (ii) AD-dementia vs non-AD dementia based on the EEG features. To evaluate the performance of the model, nested cross-validation is used with the area under the precision-recall curve (AUPRC) as evaluation metric.

Results

The binary classification of patients with dementia and controls resulted in an average AUPRC of 0.985 for the validation set and 0.965 for the test set. The classification of AD vs. other types of dementia resulted in average AUPRCs of 0.750 and 0.749 respectively for the validation and test set. Patients with dementia had significantly higher power in the delta and theta band compared to controls as can be noticed in fig.1. To classify AD vs. non-AD dementia functional connectivity features had most discriminative power.

Conclusions

EEG biomarkers can differentiate between dementia and healthy controls with excellent accuracy. Furthermore, the differentiation between AD and non-AD dementia is possible with a good accuracy, indicating the potential of EEG biomarkers to be used in clinical practice.

Aims

Impairment in network connectivity explored by resting state EEG (rsEEG) recordings, was reported in posterior cortical areas of Alzheimer’s disease (AD) patients. In the present longitudinal multicenter study we aimed to: 1. investigate whether disruption of cortical network connectivity is differently expressed in Frontotemporal Dementia (FTD) vs. AD, 2. to longitudinally characterize patterns of connectivity along the course of the disease.

Methods

Network connectivity was detected in rsEEG by Mutual Information (MI) Analysis, a measure of intererelatedness between nodes of a network. MI provided an input for graph theory analysis of the network. rsEEG was recorded in 18 FTD, 18 AD, and 20 healthy controls (HC). FTD and AD patients were recorded at the prodromal stage of dementia, at onset of dementia and three years after dementia onset. HC underwent three EEG recordings at the same interval.

Results

FTD and AD patients showed greater MI than HC at the prodromal stage of dementia (p=4·10^-7 for FTD vs. HC; 7·10^-3 for AD vs. HC). The main hub of HC (P3) was lost in FTD patients at the onset of dementia and substituted by provincial hubs in frontal leads. FTD network appeared to be rearranged in new small worlds. No changes in global network organization was found in AD.

Conclusions

Cortical neural hyperconnectivity was present only in the prodromal stage of dementia in areas involved in the specific pathological process of FTD (anterior) and AD (posterior regions), thus suggesting that it is an early electrophysiological feature, potentially useful to identify prodromal FTD and AD.

Aims

Covid-19 has accelerated adoption of “virtual” trial designs, but brain imaging still requires visits to clinical centres. The BrainWaveBank platform enables repeated recording of resting and domain-specific neurological function in homes and small clinics. Here we assess user compliance and data quality risks associated with remote neurophysiological studies.

Methods

Data from three studies is reported:

1. After in-person familiarisation, 90 adults (40-80yrs) were asked to complete 5 x 20min weekly sessions, with passive and task-driven dry-EEG, over 12 weeks.

2. Thirty young adults completed a similar task design in matched lab and home sessions.

3. Twenty young adults completed home-based sessions, with half receiving in-person training, and half video-conference based training.

Results

1. Reported acceptance (System Usability Scale) was high (78.9), but somewhat lower for users over 67yrs (68.6). Daily adherence was highest (90%) for those over 67yrs. Home-based grand-average ERP topography and morphology was similar to the literature and to in-lab recordings.

2. To yield data with similar statistical power to lab-based recordings, between 20% and 50% more home-based is needed.

3. Initial analysis showed similar noise levels and higher adherence for remotely-enrolled participants.

Conclusions

Adherence and judged usability were high, even with older individuals performing repeated unsupervised sessions. Resting state and task-synchronised EEG can be collected in the home, with modest effects on grand-average data, which can be mitigated by asking patients to do a repeat recording. Initial data does not suggest that remote familiarisation sessions increase noise or decrease adherence.

Aims

We studied retinal function using electroretinography (ERG) in preclinical Alzheimerʼs Disease (AD) classified by cerebral spinal fluid (CSF) β-amyloid/Tau ratio.

Methods

As part of an ongoing longitudinal study, we enrolled cognitively healthy (CH) individuals with normal retinal morphology by spectral-domain optical coherence tomography. CH participants were classified into two sub-groups, normal (CH-NAT) or pathological (CH-PAT), using a logistic regression model from the CSF β-amyloid/Tau ratio that identified >85% of patients with clinically probable AD. Full-field ERG (ffERG), pattern ERG (PERG), and photopic negative response (PhNR) were performed in 15 CH-PAT (30 eyes) and 14 CH-NAT (28 eyes) adult participants. Amplitude and implicit times of the ERG wave components were analyzed. Group differences were tested using mixed model repeated measures and a classification model was derived using repeated measure multivariable logistic regression.

Results

Mean age (+/- standard deviation) in the CH-PAT group (76.5 +/- 6.6 years) was similar to the CH-NAT group (75.9 +/- 8.5 years; p=0.5). Preclinical AD participants showed marked RGC dysfunction relative to controls. The PhNR was significantly diminished in preclinical AD relative to controls (~50%reduction; p = 0.003). PhNR amplitude and N95 implicit time classified CH-NAT vs. CH-PAT with 87% sensitivity and 82% specificity.

Conclusions

ERG showed marked RGC dysfunction in presymptomatic AD participants with structurally normal OCTs. Retinal functional data differentiated CH individuals with CSF biomarkers of AD pathology with high sensitivity and specificity. Retinal dysfunction measured by ERG may serve as an objective and non-invasive clinical marker for predicting AD pathology prior to detectable cognitive deficits.