Fig. 4
From: DeepNeuron: an open deep learning toolbox for neuron tracing
The workflow of neurite connection module. (a) In the training step, the connectivity of a signal pair is learned using a revised Siamese network. In each pair, two 1 × 200 feature vectors are extracted. (b) The validated result with sorted dissimilarity scores for all signal pairs shows that the dissimilarity scores of positive pairs are much lower than that of negative pairs. (c) The trained model is applied in the connection step to calculate the dissimilarity for each detected signal pair. Results of our DMST connection and the original MST connection (using distance as the weight only) are shown