Jason Kutarnia’s work deals with the challenge of creating training material for the ultrasound training simulator under development in the Ultrasound Lab, with obstetrics as the chosen demonstration clinical discipline. The training material is a 3D image volume from which the ultrasound images shown on the training simulator can be extracted.
In order that the proper image be displayed for any position of the sham (= pretend) ultrasound transducer on the ‘phantom belly,’ we need to acquire ultrasound image material from the whole belly region of pregnant subjects. That, in and of itself, is not the difficult part; we have a sonographer carry out multiple overlapping 3D scans using a transducer with a tracking sensor attached. The challenge lies in merging these image volumes into one composite image volume that is anatomically and ultrasonically correct when the anatomical structures spanning more than one volume do not line up, due to breathing articles, change in muscle tone, varying probe pressure and fetal movements.
Jason’s work deals with all of the above, but the poster presentation that received recognition at the SPIE Medical Imaging Conference dealt specifically with comparing the performance of two methods, both developed by Jason, for finding the cutting surfaces (‘seams’) which permits the morphing several 3D ultrasound image volume into one volume; the technical term for this is 3D mosaicing.
The first method models the seam as a B-spline surface and treats its calculation as a shape optimization problem, which is solved using a cooperatively coevolving particle swarm based approach. The second method treats the seam selection as a voxel labeling problem, where each voxel in the composite volume is labeled with its respective source volume. The formulation is optimized using graphcuts, which guarantees that a global minimum is achieved.