Automated registration of 3D ultrasound and CT/MR images for liver

In clinical diagnosis of liver cancer or treatment using US image guided navigation often required correlation of suspicious region in US with the clinical findings on CT or MR images. We proposed a faster automated registration technique consists of two-steps registration framework which contains vessel structure-based coarse registration and gradient feature-based fine registration. Results were validated using 30 clinical image data-sets collected in standard clinical workflow of the patient undergoing liver cancer treatment. Registration success rate found to be 73%. Registration error were measured 3.88±1.38[mm] and 3.22±1.91[mm] for US-MR and US-CT cases respectively. Total computation time was determined as 40 sec. on GE's standard Advantage Workstation.