EPV 14. Non-invasive single-trial detection of human population spike responses in somatosensory evoked potentials in a realistic clinical setting

详细信息    查看全文

• 作者：G. Waterstraat^a ; ; G. Curio^a ; ^b
• 刊名：Clinical Neurophysiology
• 出版年：2016
• 出版时间：September 2016
• 年：2016
• 卷：127
• 期：9
• 页码：e219
• 全文大小：39 K

文摘

High-frequency EEG oscillations (≈$\approx$600 Hz; HFO) evoked by median nerve stimulation and recorded above the human somatosensory cortex are non-invasive correlates of cortical population spikes. Recently, it was shown that spatiotemporal filtering and multivariate classification enables single-trial HFO detection using 29-channel low-impedance (<1 kΩ)$\mathrm{\Omega })$ low-noise EEG in an electromagnetically shielded recording chamber. It is an open question, whether this can be achieved in a realistic clinical setting.

Methods

With a custom-built CE-certified low-noise EEG amplifier, median nerve SEPs were recorded in 10 healthy subjects using 8 electrodes (impedances ≈$\approx$1 kΩ)$\mathrm{\Omega })$ in a standard unshielded hospital environment. After band-pass filtering (500–900 Hz), a subset of the trials (N = 2000) was used to train the two-step single-trial HFO detector, which is composed of spatiotemporal filter optimization and nonlinear classification. The performance of the algorithm was assessed using an independent set of additional trials (N = 5200).

Results

In the present group of 10 subjects, on average the algorithm detected evoked HFOs in 64.9% of the single trials in the correct latency window (around ≈$\approx$20 ms) with a positive predictive value (PPV) of 61.9%. Notably, in several subjects with a higher signal-plus-noise-to-noise ratio (SNNR), detection rate (DR) and PPV were above 80% (peak values: SNNR = 2.0, DR = 95.2%, PPV = 98.5%).

Conclusions

A non-invasive single-trial detection of human population spike responses in somatosensory evoked potentials can be achieved also in a realistic unshielded clinical setting. The increase in sensitivity brought about by combined hardware and algorithmic improvements enables the analysis of single-trial variability and might be extended also to pathological components, such as the non-invasive detection of epileptic neocortical high-frequency oscillations.