The invention belongs to a clustering method of
brain white matter fibers, and particularly relates to a tamp-. The invention discloses a
brain white matter fiber deep clustering method guided by fMRI. Task-is used for carrying out deep clustering on
brain white matter fibers. FMRI data are used for obtaining functional information of the
white matter fibers, meanwhile, DTI data are used for obtaining structural information of the
white matter fibers, the functional information and the structural information are combined to represent one
white matter fiber, and the white matter
fiber serves asinput of the embedded clustering
convolution automatic
encoder to be clustered. According to the method disclosed by the invention, the average tag-on the fiber track is extracted; the fMRI
signal represents a fiber, and the result of the clustering is further structurally restricted and optimized in combination with the structure information from the DTI data. Jointly inputting the two types ofinformation into a CAEEC (convolutional
autoencoder) of an embedded cluster to generate a clustered
fiber bundle, and in the CAEC training process, optimizing the training process by combining reconstruction-oriented loss, clustering-oriented loss and
sparse regularization items; the hierarchical structure of the fiber can be better extracted by the convolutional automatic
encoder embedded with the cluster, and the local features of the data are reserved in the feature space.