Michael von Rhein

Dipl.Inf. Univ. Michael von Rhein MedTec? & Science GmbH? Head of Software Development

Diploma Thesis

3D Freehand Ultrasonography using Inertial Sensors (in cooperation with Siemens Corporate Research Princeton, NJ, USA)

Supervisor: Dr. Oliver Kutter, Dr. Ali Kamen, Athanasios Karamalis M.Sc.

Abstract:
Medical ultrasound has proven its use over the last decades. It is fast, cheap, easy to use and does no harm to the patient. The low image quality of ultrasound images (US) has been improved over the time by methods like beam-forming, multi-frequency imaging and post processing image quality enhancement filters. In combination with other modalities such as computer tomography (CT) or magnetic resonance imaging (MRI) it is possible to achieve a higher level of information. We can get a better context of the anatomy or display functionalities that can only be measured by one modality in a combined image. This thesis proposes the use of inertial tracking sensors for US tracking which is used for 3D freehand ultrasound recording and registration. We especially want to remove expensive and hard to maintain tracking devices by replacing them with inertial sensors while doing a multi-modal registration with a CT volume. The benefits of this method are that the physicians can do a normal ultrasound scan with a transducer that has embedded and very easy to use sensors, while getting a 3D volume of the ultrasound scan that is already registered to another imaging modality. The main problem with inertial sensors is the increasing noise on the measured signal called drift which grows linearly on the gyroscopic devices and exponentially on the accelerometers. We introduce a concept of a multiphase sliding window optimization to remove the drift from the tracking signal and perform experiments with synthetic data to show the general robustness and fitness of our algorithm on the problem. No real inertial sensors were used in this thesis. A second problem with conventional tracking devices is the need for an initial calibration. In our approach we can remove the calibration completely by doing it implicitly in the registration. Our experiments show that we can achieve a good registration in acceptable time. A challenge for the future is the registration of image data with little details or poor quality as our method does not work perfectly with this data.

Work

Favorite Links