State-of-the-art performance in electroencephalography (EEG) decoding tasks is currently often achieved with either Deep-Learning (DL) or Riemannian-Geometry-based decoders (RBDs). Recently, there is growing interest in Deep Riemannian Networks (DRNs) possibly combining the advantages of both previous classes of methods. However, there are still a range of topics where additional insight is needed to pave the way for a more widespread application of DRNs in EEG. These include architecture design questions such as network size and end-to-end ability. How these factors affect model performance has not been explored. Additionally, it is not clear how the data within these networks are transformed, and whether this would correlate with traditional EEG decoding. Our study aims to lay the groundwork in the area of these topics through the analysis of DRNs for EEG with a wide range of hyperparameters. Networks were tested on five public EEG datasets and compared with state-of-the-art ConvNets. Here, we propose end-to-end EEG SPDNet (EE(G)-SPDNet), and we show that this wide, end-to-end DRN can outperform the ConvNets, and in doing so use physiologically plausible frequency regions. We also show that the end-to-end approach learns more complex filters than traditional bandpass filters targeting the classical alpha, beta, and gamma frequency bands of the EEG, and that performance can benefit from channel-specific filtering approaches. Additionally, architectural analysis revealed areas for further improvement due to the possible under utilisation of Riemannian specific information throughout the network. Our study, thus, shows how to design and train DRNs to infer task-related information from the raw EEG without the need of handcrafted filterbanks and highlights the potential of end-to-end DRNs such as EE(G)-SPDNet for high-performance EEG decoding.

The brain’s biological age has been considered as a promising candidate for a neurologically significant biomarker. However, recent results based on longitudinal magnetic resonance imaging (MRI) data have raised questions on its interpretation. A central question is whether an increased biological age of the brain is indicative of brain pathology and if changes in brain age correlate with diagnosed pathology (state hypothesis). Alternatively, could the discrepancy in brain age be a stable characteristic unique to each individual (trait hypothesis)? To address this question, we present a comprehensive study on brain aging based on clinical Electroencephalography (EEG), which is complementary to previous MRI-based investigations. We apply a state-of-the-art temporal convolutional network (TCN) to the task of age regression. We train on recordings of the Temple University Hospital EEG Corpus (TUEG) explicitly labeled as non-pathological and evaluate on recordings of subjects with non-pathological as well as pathological recordings, both with examinations at a single point in time TUH Abnormal EEG Corpus (TUAB) and repeated examinations over time. Therefore, we created four novel subsets of TUEG that include subjects with multiple recordings: repeated non-pathological (RNP): all labeled non-pathological; repeated pathological (RP): all labeled pathological; transition non-patholoigical pathological (TNPP): at least one recording labeled non-pathological followed by at least one recording labeled pathological; and transition pathological non-pathological (TPNP): similar to TNPP but with opposing transition (first pathological and then non-pathological). The results show that our TCN reaches state-of-the-art performance in age decoding on non-pathological subjects of TUAB with a mean absolute error of 6.6 years and an R 2 score of 0.73. Our extensive analyses demonstrate that the model underestimates the age of non-pathological and pathological subjects, the latter significantly (-1 and -5 years, paired t-test, p = 0.18 and p = 6.6 e − 3 ). Furthermore, there exist significant differences in average brain age gap between non-pathological and pathological subjects both with single examinations (TUAB) and repeated examinations (RNP vs. RP) (-4 and -7.48 years, permutation test, p = 1.63 e − 2 and p = 1 e − 5 ). We find mixed results regarding the significance of pathology classification based on the brain age gap biomarker. While it is indicative of pathological EEG in datasets TUAB and RNP versus RP (61.12% and 60.80% BACC, permutation test, p = 1.32 e − 3 and p = 1 e − 5 ), it is not indicative in TNPP and TPNP (44.74% and 47.79% BACC, permutation test, p = 0.086 and p = 0.483 ). Additionally, all of these classification scores are clearly inferior to the ones obtained from direct EEG pathology classification at 86% BACC and higher. Furthermore, we could not find evidence that a change of EEG pathology status within subjects relates to a significant change in brain age gap in datasets TNPP and TPNP (0.46 and 1.35 years, permutation test, p = 0.825 and p = 0.43 ; and Wilcoxon-Mann-Whitney and Brunner-Munzel test, p = 0.13 ). Our findings, thus, support the trait rather than the state hypothesis for brain age estimates derived from EEG. In summary, our findings indicate that the neural underpinnings of brain age changes are likely more multifaceted than previously thought, and that taking this into account will benefit the interpretation of empirically observed brain age dynamics.