Reliability and agreement of manual and automatic methods for single trial detection and estimation of somatosensory evoked potential features

J.A. BIURRUN MANRESA; F.G. ARGUISSAIN; D.E. MEDINA REDONDO; C.D. MØRCH; O.K. ANDERSEN

San Diego

Conferencia; 6th International IEEE EMBS Neural Engineering Conference (NER); 2013

SEVERAL methods have been proposed in the past for automatic detection and estimation of somatosensory evoked potential (SEP) features [1]. However, the agreement (presence/absence of SEP features) and reliability (measurement error) of these methods compared to the current gold standard (manual detection) has not been properly established. Thus, the aim of this study is to assess the agreement and reliability of manual and automatic single trial feature detection methods.SEPs were recorded from the vertex of 15 healthy volunteers, in response to electrical stimulation of the plantar nerve. A total of 120 trials were recorded for each subject. Selected features (amplitudes and latencies from N1, N2 and P2 peaks) were extracted from each trial, using three methods: manual detection, an automatic algorithm based on the derivative of the signal and classification of features using fuzzy logic (Fuzzy-Deriv), and another automatic algorithm based on wavelet filtering and multiple linear regression (WF-MLR) [1]. Agreement was assessed as overall agreement (po), agreement by chance (pe), positive agreement (ppos), negative agreement (pneg) and Cohen?s κ [2]. Absolute and relative reliability were assessed by the coefficient of variation (CV) and the intraclass correlation coefficient (ICC), respectively [3].Results are shown in Table 1. In general, there is low agreement between methods (low κ), since most of the agreement appears to happen by chance (pe is very close to po). Regarding reliability, there is a wide variation across the different features and methods, although in general variation is smaller in latencies compared to amplitudes.In conclusion, these results indicate that the interpretation of SEP features must be performed cautiously, since the outcome features (peak amplitude and latencies) are strongly dependent on the detection method used.