PyHFO: lightweight deep learning-powered end-to-end high-frequency oscillations analysis application.

Area of Science:

• Neuroscience
• Computational Biology
• Medical Technology

Background:

• Epileptogenic zone detection from EEG is crucial for epilepsy treatment.
• Traditional methods for analyzing high-frequency oscillations (HFOs) in EEG are computationally intensive.
• Deep learning (DL) offers potential for improved HFO detection but faces computational challenges.

Purpose of the Study:

• To develop and validate PyHFO, an end-to-end software platform for streamlined DL-based HFO detection in EEG.
• To enable efficient identification of neurophysiological biomarkers for epileptogenic zones.
• To provide a user-friendly and computationally efficient tool for epilepsy research and clinical practice.

Main Methods:

• Introduced PyHFO, incorporating DL models for artifact and HFO classification.
• Integrated time-efficient HFO detection algorithms (e.g., short-term energy, MNI detectors).
• Validated PyHFO on diverse EEG datasets (grid/strip, depth electrodes, rodent studies).

Main Results:

• PyHFO demonstrated efficient handling of various EEG datasets.
• Optimization techniques resulted in speeds up to 50 times faster than traditional HFO detection.
• Users can utilize a pre-trained DL model or train custom models with their own EEG data.

Conclusions:

• PyHFO effectively addresses the computational challenges of applying DL to EEG data analysis in epilepsy.
• The platform offers a feasible and efficient solution for both clinical and research settings.
• PyHFO promotes wider adoption of advanced EEG analysis tools and facilitates large-scale research collaborations.