2D neuron detection and 3D merging results.

During detection, two different sets of predefined anchor box sizes—“9 × 9” or {“7 × 7”, “11 × 11”}—are used. Three experimental 3D merging results are presented: merging of human annotated neuronal regions, where 2D detection performance is irrelevant and thereby represented by “-”; 3D merging of neuronal regions detected by an ANN. Box sizes of “7, 11” have slightly better 2D detection and 3D merging F1 score, but this combination exhibits lower inference speed and requires significantly longer training time than box size “9” does. “ms/vol” refers to milliseconds per volume.