Ruprecht-Karls-Universität Heidelberg


Three slices out of an image stack, and the color-coded dominant local orientation


Collagen, the most abundant protein in the human body, is the major component in connective tissue like the cornea. Non-invasive, high resolution imaging of collagen fibre networks is important for cornea-related biomedical research. Typical histological investigations include invasive sample preparation such as slicing, fixation and staining. With a multi-photon laser scanning microscope, the intrinsic second harmonic generation from collagen enables a novel non-invasive all-optical imaging method. Without slicing and staining, an in vivo 3D cornea imaging is possible with the advantages of strong contrast, high resolution (<1 µm), and large sensing depth (> 1 mm).

The aim of the project is to provide a structural characterization of the cornea.


Starting from the three-dimensional multi-photon microscopic images, two approaches are conceivable:
  • The blurring effect of the point spread function can be partially cancelled using deconvolution techniques. In the resultant three-dimensional image stack, multiple orientations of the overlapping lamellar structures should be estimated and segmented. The segmented lamellae should finally be visualized using isosurfaces.
  • A computationally much more demanding alternative is a Bayesian image reconstruction resulting in an explicit representation of the fibres. A model consisting of multiple chains of ellipsoidal units, developed in D.W. Heermann's lab, should be sampled to find configurations satisfying both prior knowledge (lamellar sheets are smooth) and the microscopic observations. Sampling is by methods such as Monte Carlo or simulated annealing. The charm of this approach is that image reconstruction and a basic representation are obtained in one framework. The resulting fibre network can be analyzed in terms of its topology or mechanical stability.

Parts of this project can be tackled as part of a practical or a diploma thesis. Please contact Fred Hamprecht or Meng Han for details.


Last update: 06.10.2010, 12:25
zum Seitenanfang