**5. Conclusions**

Some types of neural networks, in particular LSTMs, may be considered for estimating the directed connectivity of reconstructed EEG Sources. However, no method is flawless, and we showed that even an established method such as TRGC can generate faulty estimates. This calls for further developments. There is much potential for a hybrid approach, in which a neural network could be used as a preprocessing step to chart the

interesting directed connectivity patterns, after which a conventional method is applied for estimating them.

**Author Contributions:** Conceptualization, M.M.V.H., A.F. and I.V.; methodology, I.V. and A.F.; formal analysis, I.V.; resources (scripts), A.F. and I.V.; original draft preparation, I.V.; writing, review and editing, A.F and M.M.V.H.; visualization, I.V.; supervision, A.F. and M.M.V.H.; project administration, A.F. All authors have read and agreed to the published version of the manuscript.

**Funding:** I.V. performed this work as part of her Master in Artificial Intelligence thesis at KU Leuven. A.F. is supported by a grant from the Belgian Fund for Scientific Research—Flanders (FWO 1157019N). M.M.V.H. is supported by research grants received from the European Union's Horizon 2020 research and innovation programme under grant agreement No. 857375, the special research fund of the KU Leuven (C24/18/098), the Belgian Fund for Scientific Research—Flanders (G0A4118N, G0A4321N, G0C1522N), and the Hercules Foundation (AKUL 043).

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **Appendix A**

**Table A1.** Scheirer–Ray–Hare Test using Sensitivity as a criterion. Alpha level = 0.05. Significant results shown in bold. Models: Conv2D, TS = 2, kernel size 4 × 2 versus 4 × 4, Conv2D, TS = 3, kernel size 4 × 3 versus 4 × 4, with TS denoting the amount of time series. Model (*H* (3,32) = 16.04, *p* = 0.001) and dipole condition (*H* (1,32) = 4.53, *p* = 0.03) have significant effects on Sensitivity. Post hoc Mann–Whitney U tests (Bonferroni-corrected alpha level = 0.008) revealed that Sensitivity was significantly higher for Conv2D, TS=2(*Mdn* = 0.67, 0.67), versus Conv2D, TS = 3 (*Mdn* = 0.00, 0.17), *p* = 0.006, *p* = 0.007 for differing kernel sizes.


#### **References**

