Automatic segmentation of the optic radiation using DTI in glaucoma patients

An automated segmentation system of the optic radiation using diffusion tensor imaging (DTI) is proposed. The DTI-data is interpolated in the Log-Euclidean framework to avoid the swelling effect and regularized by diffusion filtering. Based on physiological and anatomical information, robust initial estimates of the optic radiation and the midbrain are obtained using thresholding and connectivity analysis. The estimated optic radiation initializes a statistical level set framework. The optic radiation is segmented by evolving the level set function. The segmentation is refined using the relative position between the optic radiation and the midbrain. The system is tested using eighteen DTI-datasets of glaucoma and normal subjects. The segmentation results were compared to the manual segmentation by a medical expert and found to be in agreement with the known anatomy with 83% accuracy. The automation eliminates the necessity of medical experts’ intervention to identify the optic radiation and facilitates future glaucoma studies.