Fig. 4 Interval-level confusion matrices for STPD detection on simulated spike trains. Rows denote simulator reference labels and columns denote STPD predicted labels after mapping burst, possible-burst and long-burst outputs into a single burst-family category. A shows the fully automatic threshold configuration, and B shows the simulator-informed threshold configuration, which is an upper-bound sensitivity analysis rather than a deployable real-data setting. Cell values show row-normalized percentages with interval counts in parentheses. Because STPD has no native Noisy output class, simulator Noisy intervals are evaluated as unclassified/non-pattern intervals. HF tonic was not emitted in the simulator-informed run, highlighting class-arbitration limits among high-frequency regimes. HF, high-frequency.
