Reinhard R. Beichel is an Assistant Professor at the University of Iowa with a joint appointment in the departments of Internal Medicine and Electrical & Computer Engineering. His research is focused on computer vision and graphics, aiming at medical applications. In particular, research areas include:

• medical image analysis,
• model-based segmentation methods (e.g., Active Appearance Models),
• robust segmentation algorithms (e.g., "Robust Active Appearance Models"; see Figs. 1 and 2),
• time efficient interactive segmentation refinement methods,
• virtual surgical planning systems based on Virtual Reality (e.g., virtual liver resection planning; see Fig. 3), and
• computer aided surgery/interventions.

He received his Ph.D. degree in Telematics (Computer Science) with distinction from the Graz University of Technology, Austria.
From 2000 to 2006 he was with the Institute for Computer Graphics and Vision (ICG) at Graz University of Technology, Austria, leading an interdisciplinary team working on the development of a virtual liver surgery planning system called "Liverplanner".

Videos
Eurographics 2004 (15 MB)
ECR 2004 (50 MB)
Eurographics 2003 (97 MB)





Fig. 1: Comparison of Active Appearance Model (AAM) and Robust Active Appearance Model (RAAM) matching on
a proximal phalanx X-ray image with implants. The AAM is severely influenced by the changed object appearance and
fails to deliver an acceptable result. The RAAM does not show such a behavior. In case of the RAAM, segmentation
errors mainly occur in the region of the joint which is affected by rheumatoid arthritis. Visualization of the selected
pixels during RAAM matching (in middle of Fig. 1) shows, that gray-value disturbances (implants) have not been used
to update model parameters (black pixels). This information might be useful for further analysis steps after RAAM
segmentation.




Fig. 2: Segmentation result of AAM and RAAM matching on a proximal phalanx X-ray image of the small
finger with missing information (black area). The relative overlap error was 45.2% and 3.3% for AAM and RAAM,
respectively. The area with missing information was excluded for calculating the relative overlap error. The RAAM
segmented the imaged part of the proximal phalanx well and provides a plausible estimate of object shape in the not
imaged area, whereas the AAM totally failed to adapt to available information.




Fig. 3: Virtual Reality based liver surgery planning in action. The image was
captured by a tracked camera and overlaid with the information about the virtual
objects placed in space to visualize the virtual planning process.




Contact him at reinhard-beichel@uiowa.edu.