Automated search for information content in fluorescence microscopy images using modified autofocusing approach
| Authors | |
|---|---|
| Year of publication | 2009 |
| Type | Conference abstract |
| MU Faculty or unit | |
| Citation | |
| Description | This contribution shows how to modify classical autofocusing in optical microscopy to perform automated search for information content in image data generated by the instrument. The new approach is useful for finding 3D regions of interest prior to image acquisition. This enables the user to record precisely those places in space that contain useful image data, which results in shorter acquisition time and less memory consumption as compared to the acquisition of classical large rectangular volumes (parallelepipeds) containing more background regions than object regions (especially for a sparse specimen and camera-based systems). The standard process of image focusing uses the following approach: the algorithm steps along the z-axis and evaluates focus values of each acquired image according to a focus function in order to find the maximum of this function. Unimodal behavior of the focus function is assumed, i.e. it is expected that only one important maximum exists. In real world, however, this assumption is often not fulfilled: due to thickness of the objects, their placement on the slide, and possible bending or tilt of the slide, the objects normally do not lie in one plane. There may be interesting planes of focus present in the focus function plot, which are suppressed by other, more distinctive, planes that prevent the autofocus algorithm from detecting them. Therefore, the goal of the developed application is to find all the z-planes with rich in-focus information content (e.g., central planes of cells in 3D) for a given lateral region. An adaptive algorithm (using the Normalized Variance focus function) has been developed for this purpose and is aimed especially at such non-unimodal cases. This work shows how to find also those interesting z-planes that are hard to detect even by the human eye. The technique is based on dividing the field of view into several sub-fields and applying the focus function to each of them independently. The separated results are then merged in order to gain a global view of the 3D regions of interest. In such a way, one can make use of automated focusing to obtain results that are not only more time and memory efficient, but also more accurate and reliable as compared to those of manual focusing or standard unimodal autofocusing. |
| Related projects: |