In this repository you can find transmission electron microscopy (EM) images used to train and test an algorithm for the automated detection of synaptic vesicles. The algorithm was developed, and therefore works at best, on transmission EM images. The algorithm is available in the following GitHub repository: https://github.com/Imbrosci/synaptic-vesicles-detection. The images in this repository were used in a scientific publication, currently under revision, describing the algorithm and analysing extensively its performance (Imbrosci et al., under revision at eNeuro). Briefly, the algorithm uses two consecutive classifiers based on convolutional neural networks and afterward, a threshold-based segmentation, a connected-component labelling and k-means clustering algorithm. The first 50 images were used to generate small patches, either containing (pos) or not containing (neg) a vesicle in the middle. These patches were used to train and test the two classifiers. The patches to train and test the first classifier and the second (also called refinement classifier) can be found in the patches.zip file. Concerning the patches generation, it is worth noting that in some images, a black mask was used to cover the regions outside the synaptic terminal. The reason for using this mask was that the labelling of the synaptic vesicles was often limited to the synaptic terminal while vesicles were present in the images outside this structures as well. For this reason, some patches, originated from the border of a synaptic terminal, may have a black portion. The tiny white border visible in some patches is likely originated during the saving procedure. The remaining 37 images were used to test the performance of the algorithm in its entirity. In the metadata file, it is possible to find additional information regarding the images and the patches.
