The role of visual inspection in SEMG input data validation

Abstract

Surface electromyography is widely applied in clinical and human–computer interaction contexts, but signal quality is often degraded by noise. Automated quality assessment methods exist, yet adoption remains limited due to a lack of validated benchmarks. Visual inspection is widely used but has not been systematically evaluated for reliability or validity. This study assessed visual inspection by collecting a dataset with controlled noise sources and analyzing inter-rater reliability and alignment with ground-truth labels. Raters reached majority agreement on more than 90% of samples, with Fleiss’ Kappa improving from fair in refined categories to substantial under broader schemes. Validity was strong under a simplified noise/noise-free classification, with all metrics above 0.9 and Cohen’s Kappa of 0.8775 under the Binary-strict scheme. These findings suggest majority-based visual inspection provides a reliable ground truth for distinguishing noisy from noise-free signals. A web-based tool was developed and made publicly available to support research and crowdsourced quality ratings.