Search

In this paper we discuss how the classical geodesic active contours (GAC) model is enhanced by incorporating `prior' information into the scheme. The modified model is applied to biomedical imagery, specifically serial ultrathin electron microscopy sections. The approach used is to apply prior analysis on a training set of data and provide geometric information about the target object during the process of curve evolution. The experimental results and analysis for both synthetic and real images show that the approach performs better than our previous method. It can be implemented semi-automated fashion giving significant improvements compared to a manual approach.

Real-time visualization plays an important role in medical imaging applications such as image-guided surgery, surgical simulation and medical education. For the real-time visualization and the simultaneous overlay of 3D triangulated objects onto live recordings of patient data gathered with fast imaging devices (e.g. from the beating heart) it is essential that the number of triangles used for displaying the meshes are reasonably low. For this purpose, we developed a novel, optimized surface reconstruction algorithm for planar cross sections based on the non-constrained 3D Delaunay triangulation using pre-computed contour data. The proposed method generates 50% fewer triangles for the cardiac images used in our application domain in comparison to standard reconstruction methods from planar cross sections. Besides reducing the number of triangles, our method eliminates 'stair-case' artifacts that can occur between consecutive cross sections for non-uniformly sampled input image data often found in biomedical imaging applications.

Bigot and co-workers have shown that it is possible to estimate velocities of labelled leukocytes by recording their fluorescent response across a 1D line placed orthogonal to flow in a vessel using fast line-scanning in a confocal microscope. However, as ground truth data cannot be recorded, validation of this method is not possible. In this paper, we address this issue by modelling the imaging process and by simulating virtual xt-scans. This approach yields realistic xt-scans for which velocities and positions of particles are known and therefore the performance of the previously proposed particle tracer can be tested and validated.

Analyzing dynamic biological systems, such as blood vessel growth in healing wounds or tumour development, requires high spatial and temporal resolution. Intravital fluorescence microscopy allows for longitudinal subcellular imaging, but it requires the use of advanced image analysis tools in order to quantitatively extract the relevant parameters or the topology of the underlying network structure to subsequently model and simulate such a system mathematically. We will present a fast and robust approach that estimates the vessel diameter with a low coefficient of error < 6% in settings that are typical for such in-vivo imaging scenarios with a low signal-to-noise ratio and often sub-optimal and uneven background illumination. The generated vessel network is geometrically cleansed for an optimal topological representation.