Segmentation of brain structures is necessary for treatment of patients suffering from Parkinson s disease and Moyamoya disease. From simple Gamma Knife radio surgery used for treatment of tumors, to complex stereo tactical procedures like Deep Brain stimulation (DBS), Pallidotomy and Thalamotomy which are used for cure in Parkinson s disease, the ultimate aim is to locate and segment brain structures as accurately as possible. This task is difficult in conventional MR images due to similar intensities of background and surrounding structures. Manual segmentation is also a cumbersome and time consuming task especially in 3D volumes. The need for accuracy of segmentation in brain structures also drives way for automatic methods. To help improve the segmentation results, an atlas-based segmentation scheme is presented where an already segmented high quality atlas MRI is non-rigidly registered to subject image. Consequently segmented data is mapped from atlas to subject domain usingthe transformation obtained by the end of registration process.