We present an automatic pipeline for spectral shape analysis of brain subcortical hippocampal structures with the aim to improve the Alzheimer’s Disease (AD) detection rate for early diagnosis. The hippocampus is previously segmented from volumetric T1-weighted Magnetic Resonance Images (MRI) and then it is modelled as a triangle mesh (Fang and Boas, Proceedings of IEEE international symposium on biomedical imaging, pp 1142–1145, 2009) on which the spectrum of the Laplace-Beltrami (LB) operator is computed via a finite element method (Lai, Computational differential geometry and intrinsic surface processing. Doctoral dissertation. University of California, 2010). A fixed number of eigenpairs is used to compute, following (Li and Ben Hamza, Multimed Syst 20(3):253–281, 2014), three different shape descriptors at each vertex of the mesh, which are the heat kernel signature (HKS), the scale-invariant heat kernel signature (SIHKS) and the wave kernel signature (WKS). Each of these descriptors is used separately in a Bag-of-Features (BoF) framework. In this preliminary study we report on the implementation of the proposed descriptors using ADNI (adni.loni.usc.edu), and DEMCAM (T1-weighted MR images acquired on a GE Healthcare Signa HDX 3T scanner) datasets. We show that the best quality of the DEMCAM dataset images have a great impact on the AD rate of detection which can reach up to 95 %. For further development of the modelling approach, local deformation analysis is also considered through a spectral segmentation of the hippocampal structure.