Detection of mitoses in embryonic epithelia using motion field analysis

Abstract

Although computer simulations indicate that mitosis may be important to the mechanics of morphogenetic movements, algorithms to identify mitoses in bright field images of embryonic epithelia have not previously been available. Here, the authors present an algorithm that identifies mitoses and their orientations based on the motion field between successive images. Within this motion field, the algorithm seeks 'mitosis motion field prototypes' characterised by convergent motion in one direction and divergent motion in the orthogonal direction, the local motions produced by the division process. The algorithm uses image processing, vector field analyses and pattern recognition to identify occurrences of this prototype and to determine its orientation. When applied to time-lapse images of gastrulation and neurulation-stage amphibian (Ambystoma mexicanum) embryos, the algorithm achieves identification accuracies of 68 and 67%, respectively and angular accuracies of the order of 308, values sufficient to assess the role of mitosis in these developmental processes.

Detecting Mitoses in Time-Lapse Images of Embryonic Epithelia Using Intensity Analysis

“…As we have noted previously, 30 that the lack of a clear angular preference for mitosis during gastrulation (Fig. 8) is consistent with the relatively isotropic fabric of the cells at that stage and with the relatively modest strain rates associated with all but the region near the blastopore.…”

“…7), as compared to the motion-based algorithm, which had a mean error of 35°. 30 More than 80% of the mitosis identified during neurulation were within 15°of their true orientations.…”

“…The data are the same as those to which the earlier motion-based algorithm was applied. 30 The gastrulation-stage sequence consisted of 25 images collected at 5-min intervals from region A in Fig. 1 and the images had 400 pixels per mm.…”

“…30 This approach works well because the intensity changes produced by mitoses are highly localized and affect only a small fraction of the total image area. Previous studies 30,36 have shown that the remaining deformation fields are smooth under normal circumstances, considerably simplifying the mapping problem, and allowing a variety of options to be considered. Here, a regular mesh (material reference frame) 24,26,34 consisting of subcellular-sized triangles is used to map corresponding areas of the image to each other.…”

“…Because cells in axolotl embryos are contiguous, have unreliable edge contrast, and fluorescent markers are not currently available for early stage embryos of this species, none of these approaches is suitable. A previous algorithm developed by the authors Siva et al 30 used image processing and pattern recognition methods to identify localized deformation patterns characteristic of mitosis, thereby avoiding the segmentation problem. That approach was able to achieve identification accuracies slightly less than 70% and average orientation errors of 27°.…”