Nuclear morphology

Alterations in nuclear morphology are an indicator of changes in nuclear mechanics which can direct gene expression. In order to measure the morphology of the nucleus, we have developed an automated, 3D image processing technique that can obtain nuclear morphology from confocal images. For this, we have extended a previously reported two-dimensional image processing algorithm [1] to three dimensions. The 2D algorithm works by identifying an approximate nuclear boundary by thresholding and then converging it to the actual boundary using an optimisation technique.

One might naively assume that we can use the 2D algorithm on each slice of the confocal stack to obtain the nuclear boundary on that slice and then stitch these together to obtain the nuclear surface. However, such a strategy does not work very well due to reasons explained next. Image processing algorithms identify objects of interest in an image using either the difference in intensity of foreground pixels in comparison to the background or by using the gradient in intensity to identify the edges. In the case of the confocal slices close to the top and bottom of the nuclei, the fluorescence signal from the nuclei is low and the signal to noise ratio is poor. Hence, there is poor contrast between the foreground (nucleus) and background pixels. Furthermore, at these slices the gradient of the image is along the z direction. This gradient information, along the z direction, is not available when we calculate the gradient on a single z slice image. Hence, a 2D code will give spurious contours, dependent on the noise in the image, at the top and bottom of the nuclei as can be seen in the Fig. 1. The 2D code we used is an implementation of the algorithm in [1]. In the slices where there are adequate signal from the nucleus (Fig. 1 third row), the 2D algorithm has converged to the actual boundary. However in slices with low signal from the nucleus (Fig. 1 second and last row) there are spurious boundaries.

Figure 1: Nuclear boundary using 2D code on each slice of a confocal stack

In order to retain the gradient information in the z direction we developed a 3D image processing code that processes the entire confocal stack together. We first obtain an approximate boundary surface by thresholding the entire stack. We then converge this approximate boundary to the actual boundary using a 3D optimisation algorithm. The video below shows the approximate nuclear surface converging to the actual surface. Details of this method are available in the supplementary information of our pre-print.

References

[1] Driscoll, M. K., Albanese, J. L., Xiong, Z.-M., Mailman, M., Losert, W., & Cao, K. (2012). Automated image analysis of nuclear shape: what can we learn from a prematurely aged cell? Aging, 4(2), 119–32