| One approach to understanding human and animal behavior rests on the visual identification of anatomical structures that relate structural patterns to development, adaptation, physiology, and function. In the study of stained brain sections, the goal is to visually identify neuro-anatomical nuclei and sub-nuclei. This identification is often made based on personal expertise in the area, and is subject to great inter-observer variability. In many instances, stained anatomical images are identified within a localized region by size, shape and distribution of elemental structures. These properties are closely linked to localized texture, which is presumably used as a measure for identifying anatomical structures. In this research, we investigated the identification of neuro-anatomical structure using a multiresolution approach based on Gabor wavelet transform. This transform was utilized because of its inherent dependence on orientation of the structural elements, its spatial localization and the fact that local texture can be extracted.; A multiresolution texture feature vector was constructed which consists of localized texture energies along different orientations and under different scales. Based on this texture energy feature, brain scan images were segmented using partitional clustering in the feature space. The characteristics of the cluster center could be viewed as a quantitative property of the underlying anatomical structure it represents. We have been able to identify the abducens and the vestibular nuclei in brainstem images. By using a revised texture feature, which uses the mean texture energy and standard deviation within each frequency band, we have been able to further segment a nucleus into sub-regions. Cluster relaxation and other performance enhancement techniques were implemented in order to make the system a potential tool for identifying, segmenting and quantifying anatomical structures. As a further demonstration of the wide range of applicability of the method in discriminating orientation, anatomical structures related to bone have been assessed as an avenue for future research. Results indicate that the texture features we defined may be important for anatomical analysis where texture is important for understanding the underlying anatomical organization. |
