in medical imaging
"Methods of Statistical Modeling in Medical Imaging"
The research group - founded in September 2008 within the Institute for
Biomedical Image Analysis (IBIA) - is dealing with the development of statistical
models of human anatomy and physiological processes. In this context the
focus of the research group is twofold:
The first focus is the development, extension and application of algorihms
and methods that allow the automatic creation of statistical models of shape
and appearance in 2, 3 and 4 (time series) dimensions. This incorporates the
following fields of (medical) image anaylsis:
The second focus is the application of the developed statistical models of shape
and appearance for diagnostic, therapeutic and interventional procedures.
This covers especially the following topics:
images (e.g. x-ray,ultrasound, CT, MR, PET, SPECT)
properties (e.g. shape, inner structure, appearance) of an organ or organ part
medical image data of patients
The basis for this research is formed by the so called "InShape" models that have
already been developed at the IBIA and were already successfully used for
the segmentation of medical images
Moreover, InShape models were used for the automatic and objective
placement of regions of interest (ROIs) and the assessment/prediction of
pathogenetic factors and biomechanical paramters (more>>)
All developed algorithms and applications are implemented using C++. The
By this means different image-processing and image-analysis modules have been
developed at the IBIA.
In the following a brief overview of these modules will be given:
This module provides an interface to selected (partly ITK) filters for image blurring,
smoothing, thresholding, morphological operations, transformations,
resampling,... in combination with 3 image viewers the are
showing an image block in 3 orthogonal directions.
Screenshot of the image preprocessing module of the I-PRESP image processing and
This module provides access to the ITK Level set toolbox. In particulary to the
fast marching level seets, active geodesic level sets and threshold level set filter.
In combination with basic manual 2D segmentation tools it provides functionalilty
for interactive semi-automatic segmentation of 3D image data.
Demo video of a segementation of the left cardiac ventricle using fast marching
and active geodesic level sets (flash player needed)
This module allows the automatic creation of 3D shape and appearance (InShape)
models based on deformation fields created.
In order to establish correspondences different combinations of rigid, affine and
elastic registration (diffeomorphic demons, B-Spline registration) can be used.
As input the user must provide image data and segmented binary labes of the
organ that shall be modeled.
Screenshot of the InShape model creation dialogue
Screenshot of the dialogue for regression model calculation and parameter prediciton
With this module the created models can directly be used for segmentation.
A combination of model based segmentation and level set segmentation is
used for the automatic segmentation of 3D (+time) medical image data.
Proximal femur segmentation in CT image using 3D InShape model+level sets (left)
and 3D segmentation result (right) (flash player needed)
3D endocardium segmentation using InShape model + level sets (flash player needed)
Using this module the shape and appearance variations represented
by different PCA modes can be visualized using different types of representations.
Moreover, user-defined model instances can be created and saved for further processing.
Demo video shwoing the model explorer module (flash player needed)
This module provides tools for 3D surface rendering
of binary data using marching cubes algorithm. The modeled scene can be
edited by changing colors, lightning, type of surface representation, texturing,...
Screenshot of the 3D visualization module
2D/3D registration module XSePT
(X-ray "Segistration" Process Toolkit)
XSePT is a generic scientific tool for model-based 2D/3D image "segistration".
It provides functionality for the (semi) automatic registration of either static 3D images
(e.g. CT, MRI) or 3D InShape models on 2D images (e.g. x-ray).
For this purpose digitally reconstructed radiographs (DRR) are computed using
techniques like ray casting and wobbled splatting.
XSePT was developed at the IBIA by DI Philipp Steininger
Screenshot of the 2D/3D "segistration" module
Demo video of 2D/3D hip registration using XSePT(flash player needed):
Contact: Karl D. Fritscher, MD, PhD Head of the Research Group Room: G3 41
Tel: +43-50-8648-3866 Fax: +43-50-8648-673866