Detecting drug-induced EEG alterations with machine learning: A methodological proof-of-concept study in rats treated...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Pharmacology

Background:

• Deep learning for electroencephalogram (EEG) analysis is advancing rapidly.
• Existing architectures may not be optimal for low-channel animal EEG recordings.
• This study explored deep learning's potential for analyzing rat EEG data.

Purpose of the Study:

• To investigate deep learning's ability to detect EEG alterations induced by the NMDAr antagonist MK-801 in rats.
• To assess the suitability of deep learning for brain circuit-related efficacy and safety evaluations.
• To evaluate deep learning performance on low-channel animal EEG data with minimal preprocessing.

Main Methods:

• EEG recordings were obtained from male Sprague-Dawley rats using telemetry implants in a crossover design.
• Signals were segmented into 0.5-s epochs and processed with minimal preprocessing.
• Several deep learning architectures were evaluated for classification tasks, including EEG Conformer.

Main Results:

• EEG Conformer achieved an F-score near 0.90 on the recording-level test set.
• Leave-one-subject-out (LOSO) cross-validation yielded a mean weighted F1-score of 0.646 ± 0.034, indicating generalization to unseen animals.
• MK-801 dose-dependently increased abnormal EEG classifications, and auditory stimulation enhanced detection sensitivity.

Conclusions:

• Deep learning can effectively detect compound-related EEG alterations in low-channel rat recordings with minimal preprocessing.
• The study demonstrated generalization to unseen treatment conditions and entirely unseen animals.
• This approach can complement conventional methods, supporting automated and standardized preclinical EEG workflows for safety and efficacy studies.