Dendritic spines of pyramidal neurons are thin protrusions of the neuron membrane and represent the main postsynaptic sites in the cerebral cortex. Spines can have different morphologies: investigating their shape gives indications about the functional state of synapses during brain development. Quantifying dendritic spine shape changes will boost the comprehension of several neuropathies, such as Alzheimer and autism spectrum. Detection of dendritic spines is challenged by optical limitations. We propose a fully automated pipeline to segment and cluster spines to classify them according to their morphology.Confocal stacks of intracellular injected layer III pyramidal neurons from the human brain cingulate cortex have been segmented to reveal the dendrite core along with their spines. Segmentations were obtained with an algorithm we developed which exploits topological information. Skeletons of such segmentations were reconstructed via NeuTube, the dendrite core was identified and separated by spines. Then, the geometry of each spine in the segmented datasets is described by a set of polygonal meshes, and, to delineate spine shape and size, a set of geometrical features have been extracted. Finally, we exploited the K-means clustering algorithm to classify the spines according to their morphological structures.The proposed automatic pipeline allows to segment and classify multiple classes of spines that matched with the different shapes known in literature.The classification of the spines based on their morphology will support the study of the cellular mechanisms underlying brain pathophysiology. The integration of other geometric features will allow to improve dendritic spine classification.