Page 127 - AN-4-2

Advanced Neurology ML for EEG signal recognition

validation curves and unstable loss trajectories) indicates References
a need for further optimization to improve generalization. 1. Guerreiro CM. Epilepsy: Is there hope? Indian J Med Res.
DNNs, though effective in some contexts, struggled to 2016;144(5):657-660.
match CNN performance due to their inability to capture
the intricate spatiotemporal dependencies inherent in EEG doi: 10.4103/ijmr.IJMR_1051_16
signals. These results align with broader methodological 2. Pothmann L, Klos C, Braganza O, et al. Altered dynamics
critiques: Studies relying solely on classification metrics of canonical feedback inhibition predicts increased
(e.g., accuracy, F1-score) without transparently evaluating burst transmission in chronic epilepsy. J Neurosci.
model dynamics (e.g., learning curve convergence, ROC 2019;39(45):8998-9012.
threshold stability) risk overestimating clinical utility. doi: 10.1523/JNEUROSCI.2594-18.2019
This work underscores that algorithmic performance 3. Fisher RS, Cross JH, D’Souza C, et al. Instruction manual
claims must be paired with rigorous validation of training for the ILAE 2017 operational classification of seizure types.
behaviors. CNNs not only achieved better accuracy but also Epilepsia. 2017;58(4):531-542.
demonstrated stable convergence patterns and balanced doi: 10.1111/epi.13671
TPR-FPR tradeoffs across thresholds, fulfilling key criteria 4. Blumenfeld H. Arousal and consciousness in focal seizures.
for clinical robustness. These findings solidify the promise Epilepsy Curr. 2021;21(5):353-359.
of CNNs as a reliable foundation for automated epilepsy
screening tools. Prioritizing architectures with verifiable doi: 10.1177/15357597211029507
generalization behaviors is necessary to bridge the gap 5. Brodovskaya A, Kapur J. Circuits generating secondarily
between computational advancements and clinically generalized seizures. Epilepsy Behav. 2019;101:106474.
deployable solutions. doi: 10.1016/j.yebeh.2019.106474

Acknowledgments 6. Benbadis SR, Beniczky S, Bertram E, MacIver S, Moshé SL.
The role of EEG in patients with suspected epilepsy. Epileptic
None. Disord. 2020;22(2):143-155.

Funding doi: 10.1684/epd.2020.1151
7. Chen H, Koubeissi MZ. Electroencephalography in epilepsy
None. evaluation. Continuum (Minneap Minn). 2019;25(2):431-453.
Conflict of interest doi: 10.1212/CON.0000000000000705
The authors declare no conflicts of interest. 8. Müller-Putz GR. Electroencephalography. Handb Clin
Neurol. 2020;168:249-262.
Author contributions doi: 10.1016/b978-0-444-63934-9.00018-4
Conceptualization: All authors 9. Kappel SL, Kidmose P. High-density ear-EEG. Annu Int
Formal analysis: All authors Conf IEEE Eng Med Biol Soc. 2017;2017:2394-2397.
Investigation: All authors doi: 10.1109/embc.2017.8037338
Methodology: Daniel Nasef, Milan Toma 10. Leibetseder A, Eisermann M, LaFrance WC Jr., Nobili L, von
Project administration: Milan Toma Oertzen TJ. How to distinguish seizures from non‐epileptic
Supervision: Milan Toma manifestations. Epileptic Disord. 2020;22(6):716-738.
Writing–original draft: All authors
Writing–review & editing: All authors doi: 10.1684/epd.2020.1234
11. Mari-Acevedo J, Yelvington K, Tatum WO. Normal EEG
Ethics approval and consent to participate variants. Handb Clin Neurol. 2019;160:143-160.

Not applicable. doi: 10.1016/b978-0-444-64032-1.00009-6
Consent for publication 12. Kubota T, Kidokoro H, Narahara S, et al. Evaluation of
interobserver variability in application of the new neonatal
Not applicable. seizure classification proposed by the ILAE Task Force.
Epilepsy Behav. 2020;111:107292.
Availability of data doi: 10.1016/j.yebeh.2020.107292

The data utilized in this study is open-source and properly 13. Wilcox KS, West PJ, Metcalf CS. The current approach of
referenced. the Epilepsy Therapy Screening Program contract site

Volume 4 Issue 2 (2025) 121 doi: 10.36922/an.7941