Abstract
M.Ing. (Electrical and Electronic Engineering)
Advances in bio-imaging have triggered the development of a highly sophisticated
imaging tool known as fluorescence microscopy. Fluorescence microscopy is used in
many biological applications to visualize sub-cellular processes and gives the ability
to image three-dimensional (3D) structures in living cells. The use of fluorescence
microscopy and specific staining methods make biological molecules appear as bright
spots in image data. The analysis of fluorescence microscopy images requires the
detection and tracking of hundreds spots in image data and is of great importance for
biologists to better understand cell functions. However, the analysis of these data is
still performed manually in most biological laboratories worldwide. The manual
analysis of these data is both time consuming, laborious and susceptible to human
errors. Several computer-based algorithms have been proposed for the detection and
tracking of spots in microscopy images. Most of these methods were validated on
limited image data and relatively few studies have been performed for the comparison
of these methods in real applications.
This study quantitatively compared the performance of four detection and two
tracking methods applied in microscopy images for the analysis of bright spots. The
performance of the algorithms was validated on both synthetic and real images. The
synthetic images offered a better way of validating algorithms against ground truth
reference results.
Results indicate that there are major differences in algorithm performance for both
detection and tracking. In the detection results the Isotropic Undecimated Wavelet
Transform (IUWT) and the Laplacian of Gaussian (LoG) achieved better results than
the other methods in comparison when values are considered. The tracking
results indicate that the Interacting Multiple Model (IMM) method achieved better
results than the Feature Point Tracking (FPT) method when Jaccard Similarity Scores
(JSC) are considered.