Figure 5.

neuTube 1.0 helps produce significantly more accurate neuron structures than Neuromantic and Neurostudio do. A, Node_g is a critical point from ground truth neuron and Node_t is a critical point from tracing result. These two points can be matched when $T_d=T_{d2}$ because the distance between them is less than $T_{d2}$. They cannot be matched when $T_d=T_{d1}$. B, The solid curves show average errors measuring the discrepancy between the critical point sets from user reconstruction and ground truth under different distance thresholds. The surrounding envelopes are the 95% confidence intervals. The error curve of neuTube 1.0 (green) is consistently lower than the other two, with p < 0.01 (t test) when $T_d\ge 6$ (compared to Neurostudio) or $T_d\ge 5$ (compared to Neuromantic).