Fig. 3.

Receiver operating characteristic (ROC) curves for FP identification (phase three). (a) Comparison of two analytical models. For each muscle (biceps in red and gastrocnemius in blue), 40 one-minute representative recordings were analysed. Model one (M1, triangle symbols) took the form Y = A₁ and model two (M2, circular symbols) took the form Y = A₂X_, where Y was the optimal AT_inc (μV), X was the mean noise band (μV) and A_1/2 was a positive value. From left-to-right, threshold values for A₁ were 100, 60, 40, 35, 30, 25, 20, 15, 10 and 5, and values for A₂ were 14, 12, 10, 9, 8, 7, 6, 5, 4 and 2. Median sensitivity and specificity are displayed (non-parametric distributions). Area-under-the curve (AUC, 2 s.f.) represents the accuracy of each model for each muscle. * indicates the performance of M1 when A₁ = 40; ∼ indicates performance of M1 when A₁ = 20; # indicates the performance of M2 when A₂ = 7. (b) Pooled results for model two. The results from biceps and gastrocnemius were pooled to produce a total of 80 one-minute recordings. Threshold values for A₂ are displayed. Curves for median sensitivity and specificity are plotted alongside lower and upper limits for 95% confidence interval (CI). AUC represents the accuracy of this model. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.).