Graph Wavelet Packets for the Classification of Brain Data in Anxiety and Depression

Recent research has been focusing on Graph Signal Processing (GSP) to combine different neuroimaging modalities, enabling the integration of both structural and functional brain data. To characterize how signals interact with brain networks, the Fourier and wavelet transforms have been extended to the graph setting by designing spectral filters on the structural graph Laplacian eigenvalues. Here, we present the benefits of leveraging graph wavelet packets in neuroimaging using diffusion MRI and fMRI data, based on the Boston Adolescent Neuroimaging of Depression and Anxiety (BANDA) dataset. We consider the distance between eigenvectors to extract features related to the spectral domain of the structural graph. Our proposed framework demonstrates superior accuracies in a classification scheme compared to conventional GSP methods.