MULTISCALE QUANTITATIVE ANALYSIS OF MICROSCOPIC IMAGES OF ICE CRYSTALS

Modern research in cryobiology requires a deeper understanding of the influence of different factors on the cryopreservation of cells, tissues, and organs. One of these factors is ice crystallization which has a tremendous impact on the surveillance and quality of live objects during freezing, long term storage, and thawing. Analysis of this process requires software which should be able to obtain quantitative parameters of crystals in a human-like manner with acceptable processing speed. The purpose of this work is to consider the possibility of using the multiscale image representation for the quantitative analysis of ice crystals. In our research, we used microscopic images of ice crystals during crystals formation and thawing. In previous studies for the segmentation of ice crystals on the image, we used different approaches such as active contour. At the same time, it should be noticed that the speed of active contour expansion is low and thus time-consuming to process large time sequence. Thus, we suggest the application of a Gaussian Pyramid. This multiscale representation allows analysis at a low scale and improves at a high scale. We have analyzed multiple images using the proposed approach. The results in the first approximation show a 2-fold increase in speed when using our implementation of active contours. At the same time, the segmented areas of crystals correspond to the approach without the use of multi-scale image representation. The results of this work show that multiscale image representation can be applied to improve the speed and applicability of modern software for automated image analysis. The next steps will include applicability evaluation of multiscale representation for different cases as well as the development of software realizing vector processor architecture.