User:Pforoughi

From CISSTwiki

Pezhman Foroughi

• Company: JHU
• Department: NSF ERC CISST
• Office phone: 1-410-516-4779
• E-Mail:
• Web page: http://cs.jhu.edu/~pezhman

Biography

I am a PhD student of Johns Hopkins University, department of Computer Science and a member of “Computer Integrated Surgical Systems and Technology (CISST)” and "the Laboratory for Computational Sensing and Robotics (LCSR)". I received my M.Sc. and B.Sc. in Electrical and Computer Engineering from Queen’s University (Kingston, Canada, 2006) and Isfahan University of Technology (Isfahan, Iran, 2004) respectively. My main research interests are medical image processing and more specifically ultrasound imaging and its application in diagnosis and image-guidance.

Projects

Ultrasound Based Localization of Pelvic Coordinate System

In Total Hip Replacement (THR) procedures, misalignment of the acetabular component can lead to dislocation and impingement. For the successful alignment of acetabular component, precise estimation of pelvic anatomical coordinate system is crucial. CT scan and fluoroscopy have been two of the common imaging modalities used to locate the anatomical coordinate. Conventional approaches include implanted bone fiducials or invasive probing of bony landmarks with a tracked pointer. In this project, we developed an ultrasound-based approach that exploits prior knowledge about the anatomy of the pelvis in the form of a three-dimensional surface atlas. We collect tracked ultrasound images from the pelvis and extract sample points. We register these points to a statistical atlas of the pelvis in which a canonical anatomical coordinate system had been defined by the surgeon, and thus localize this generic coordinate system in the specific patient.

Ultrasound Bone Segmentation

Segmentation of bone surface in ultrasound images has numerous applications in computer aided orthopedic surgery. A robust bone surface extraction technique for ultrasound images can be used to non-invasively probe the bone surface. In this work, we developed an intuitive and computationally inexpensive bone segmentation approach. The prior knowledge about the appearance of bone in ultrasound images is exploited toward achieving robust and fast bone segmentation. Continuity and smoothness of the bone surface are incorporated in a cost function, which is globally minimized using dynamic programming. The performance of this method is evaluated on ultrasound images collected from two male cadavers. The images are segmented in about half a second making the algorithm suitable for real-time applications. Comparison between manual and automatic segmentation shows an average accuracy of less than 3 pixels (0.3 mm).

Automatic Initialization for 3D Bone Registration

In image-guided bone surgery, sample points collected from the bone surface are registered to the pre-operative CT model using well-known methods such as Iterative Closest Point (ICP). Proper initialization of the surface points is critical for successful registration. The initial alignment is normally carried out manually due to high risk of the optimization algorithm getting trapped in a local minimum. In this project, we developed an automatic method that initially aligns the sample points collected from the surface of pelvis with the CT model. The main idea is to exploit a mean shape of the pelvis created from a large number of CT scans as the prior knowledge to guide the initial alignment. The CT model is first aligned with the mean shape using the bilateral symmetry of the pelvis and the similarity of multiple projections. Incorporating the information about the protocol of the data collection, the sample surface points are aligned with the pelvis mean shape. This will, in turn, lead to initial alignment of the sample points with the CT model. Our experiments using a dry pelvis and two cadavers show that the method can align the randomly dislocated datasets close enough for successful ICP registration. The standard ICP has been used for final registration of datasets.

Elastic registration of 3D Ultrasound Images

During a surgical operation, the anatomies with soft tissue such as liver are deformed by respiration, heartbeat, or ultrasound transducer pressure. This deformation causes difficulty for the surgeon to track a tumor or an organ landmark. In this work, we developed a novel, fast, elastic registration technique for both two-dimensional and three-dimensional ultrasound images. In this technique, a new feature vector (attribute vector) is defined for each point in the images, which is employed to find correspondence across the images. Furthermore, this vector helps select a number of key points (leading points) that are mostly located on the anatomical features of the images and can be registered robustly. The displacement of the rest of the points is derived using a fast, free-form warping technique. The proposed registration technique is computationally efficient since the elements of the new attribute vector are fast to be computed and no pre-segmentation or numerical optimization is required. Experiments are carried out to evaluate the proposed registration technique using ultrasound images of liver deformed both naturally and artificially. In these experiments, the effect of registration parameters on the accuracy and the speed of registration is investigated. The results show that the method is sufficiently accurate and robust even in cases where artifacts such as shadows exist in the ultrasound data. As an additional application, a new extended field-of-view algorithm is also developed, which estimates the translational displacement of the ultrasound volumes without requiring position tracking devices.

Based on the proposed elastic registration method, a temporal registration technique is also developed for registering a sequence of ultrasound images. Assuming that the images are consecutively captured in real time, this algorithm registers each image to the first image in the sequence. The speed of registration of each image is significantly improved exploiting the information from the registration of the previous image. The registration parameters that provide a good trade off between speed and accuracy are also found.

Publications

Please refer to my web page at: http://cs.jhu.edu/~pezhman