Redaction of false high frequency oscillations due to muscle artifact improves specificity to epileptic tissue

Abstract

Objective: High Frequency Oscillations (HFOs) are a promising biomarker of epilepsy. HFOs are typically acquired on intracranial electrodes, but contamination from muscle artifacts is still problematic in HFO analysis. This paper evaluates the effect of myogenic artifacts on intracranial HFO detection and how to remove them. Methods: Intracranial EEG was recorded in 31 patients. HFOs were detected for the entire recording using an automated algorithm. When available, simultaneous scalp EEG was used to identify periods of muscle artifact. Those markings were used to train an automated scalp EMG detector and an intracranial EMG detector. Specificity to epileptic tissue was evaluated by comparison with seizure onset zone and resected volume in patients with good outcome. Results: EMG artifacts are frequent and produce large numbers of false HFOs, especially in the anterior temporal lobe. The scalp and intracranial EMG detectors both had high accuracy. Removing false HFOs improved specificity to epileptic tissue. Conclusions: Evaluation of HFOs requires accounting for the effect of muscle artifact. We present two tools that effectively mitigate the effect of muscle artifact on HFOs. Significance: Removing muscle artifacts improves the specificity of HFOs to epileptic tissue. Future HFO work should account for this effect, especially when using automated algorithms or when scalp electrodes are not present.

Trends in the use of automated algorithms for the detection of high‐frequency oscillations associated with human epilepsy

“…Filtering of a sharp transient in the EEG will produce a burst of high frequency activity, which may be falsely detected as an HFO 34 . Signals of non‐neural origin, such as muscle activity and harmonics of electrical line noise, can also cause false‐positive detections 59 . For this reason, many algorithms include steps for rejection of false positives after the initial detection (Table 1).…”

Preictal variability of high‐frequency oscillation rates in refractory epilepsy

“…We also applied an additional, published artifact rejection method designed to redact activity associated with scalp muscle artifact, which can produce many false-positive detections in the lateral temporal lobes. 26 All HFOs discussed in this work were subjected to this full process.…”

“…Early work with small cohorts showed that preictal HFOs have subtle changes in the preictal period, such as spectral and rate changes 22 or alterations in HFO features. 23 Newer hardware and software now make HFO research much more robust, allowing high-quality, larger datasets 13,[23][24][25][26][27] ; the role of HFOs as a preictal biomarker can now be answered with much higher rigor. To our knowledge, there is no study of peri-ictal HFO rates using modern equipment and algorithms to acquire a robust sample size.…”

Tractor Assistant Driving Control Method Based on EEG Combined With RNN-TL Deep Learning Algorithm

“…Brainwaves associated with motor imagery are delta (0.5-4 Hz), theta (4-8 Hz), alpha (8-13 Hz), beta (13-30 Hz), and gamma (30-50 Hz) waves which could be captured and processed by Emotiv EPOC+. Also, the frequency of EMG artifact [26] caused by the sub-scalp EMG contraction is generally above 100 Hz. Therefore, the 8th order of 50 Hz Butterworth low-pass filter was used to filter out the highfrequency irrelevant noise signals and the delta, theta, alpha, beta, and gamma waves were extracted.…”