Data-Driven Corpus Callosum Parcellation Method Through Diffusion Tensor Imaging

Abstract

The corpus callosum (CC) is a set of neural fibers in the cerebral cortex, responsible for facilitating inter-hemispheric communication. The CC structural characteristics appear as an essential element for studying healthy subjects and patients diagnosed with neurodegenerative diseases. Due to its size, the CC is usually divided into smaller regions, also known as parcellation. Since there are no visible landmarks inside the structure indicating its division, CC parcellation is a challenging task and methods proposed in the literature are geometric or atlas-based. This paper proposed an automatic data-driven CC parcellation method, based on diffusion data extracted from diffusion tensor imaging that uses the Watershed transform. Experiments compared parcellation results of the proposed method with results of three other parcellation methods on a data set containing 150 images. Quantitative comparison using the Dice coefficient showed that the CC parcels given by the proposed method has a mean overlap higher than 0,9 for some parcels and lower than 0,6 for other parcels. Poor overlap results were confirmed by the statistically significant differences obtained for diffusion metrics values in each parcel, when using different parcellation methods. The proposed method was also validated by using the CC tractography and was the only study that proposed a non-geometric approach for the CC parcellation, based only on the diffusion data of each subject analyzed.