ID 355 – Comparison of automatic and visual high frequency oscillations detection

Abstract

Purpose To compare automatic and visual high frequency oscillation (HFO) detection. Methods We compared automatic and visual high frequency oscillations (HFO) detections in 5 patients. HFO were automatically detected by a custom made matlab detection algorithm. First, the signal in the statistical window (10 s) was filtered in a series of overlapping frequency bands. Power envelopes and the frequency stability between narrow band signal and broad band signal were computed. HFO detections were obtained by thresholding the normalized dot product of the power envelopes and frequency stability. To increase algorithm specificity amplitude, frequency stability and duration thresholds, previously obtained from HFOs marked by expert reviewer, were applied on putative HFO. This automatic detection was compared with visual HFO analysis. Results We reached excellent congruence between manual and automatic HFO detection measured by Cohen’s kappa ( ⊇ = 0.8 ). Conclusion We managed to prove excellent concordance between automatic and visual HFO detection.