A large margin nearest cluster metric based semisupervised clustering algorithm for brain fibers

Biomedical science has proven that human brain fiber tracts have correspondent relationship with the physiological functions, and it has important medical significance to cluster the brain fibers accurately. But because of the huge number of brain fibers, manually segmenting brain fibers will result in time and effort consuming. And for the extreme complexity of the distribution of brain fibers that different types of brain fibers cross with each other, automatically mapping brain fibers using unsupervised clustering algorithms cannot give satisfactory results. This work proposed a Large Margin Nearest Cluster metric based semi-supervised clustering algorithm called LISODATA, which can better separate crossing fiber tracts by employing a small amount of supervised information. The experimental results on the brain fiber dataset provided by the 2009 PBC demonstrated that LISODATA could improve the purity of brain fiber clusters compared to ISODATA.