Biomedical Imaging Group
EPFL   Student Projects: Jacques Neuenschwander
English only    BIG > Teaching > Student Projects > Jacques Neuenschwander
Home page
Tutorials and Reviews
Download Algorithms
bulletStudent projects
Detection of DNA-molecules in cryo-electron microscopy imagest

Jacques Neuenschwander
Section Microtechnique, EPFL

Semester project
June 21, 2002


The recent developments in biotechnology has made it necessary for the biologists to be able to observe small biological structures such as DNA, viruses, bacteria or protein molecules in detail. Derived from electon microscopy, cryo-electron microscopy is a recently developed method that offers many advantages for observing such living structures. However, the very poor quality of the images that can be obtained is a major downside of this method. In this context, image processing can be a good alternative to enhance cryo-electron microscopy images.


In this project we aim at reconstructing the boundaries of DNA-molecules from such an image (Fig. 1) automatically. For this challenging task we first use a feature detection method based on the multiscale computation of local moments. We then experiment two contour following methods for reconstructing the DNA strand. The first one consists in selecting the most relevant structures in a post-processing step (that involves a combination of non-maximum suppression and hystheresis threshold algorithms), and afterwards a basic contour follwing method is applied. In the second one, a more complex algorithm based on dynamic programming performs the contour tracking operation on a non-post-processed image.


Both methods were implemented and their efficiency confirmed on synthetic images. However, It was not possible to achieve a complete and automatic reconstruction of the DNA strand on a real image (Fig.2-3). Some human interaction may be needed to obtain a better performance.


Figure 1:
DNA-molecule in cryo-electronic microscopy image.
Figure 2:
Contour following with initial tracking method
Figure 3:
Contour following with dynamic programming-based method