Digital Images and 3-dimensional volumes.

The Human eye and brain together form an unparalleled imaging system which is able to recognise millions of colours, shapes and forms almost instantaneously. Computers and cameras (or photomultipliers) by comparison are really very poor and while cameras may be capable of capturing images at high resolution, once digitised the image will be typically reduced to 256 colours or grey levels for the purpose of image processing. All of the images contained within this web site are 8 bit and fall into this category. High end systems now work with 24 bit images and in time this will improve further.
Above. 3-dimensional reconstruction of raw and processed data from a segment of myograph mounted (pressurised) rabbit cutaneous resistance artery. The figure shows identical views of the same data set after different rendering and analysis processes. a) extended focus view of the nuclei within the wall of the vessel. Note that some nuclei are brighter than others (object intensity heterogeneity.). b) The volume has been rendered using a back-to-front (BTF) algorithm which allows for the control of opacity of individual voxels. In this case the background (i.e. darkest) voxels have been made transparent. This method is effective for volume examination and identification of potential problems. One particular group of nuclei were found to be touching (object fusion) and have been circled in the figure. c) The data volume was them processed using the IMTS routine (without object classifiers) which segmented the objects. Some individual nuclei were clearly segmented while some others were not. Some degree of object fracturing and fusion was observed. d) The segmented volume was then passed for automated analysis. Only the longest cord of each object is shown. These cords are 3D-vectors which describe the length and orientation of the object. It is clear that the software has failed to identify certain ‘individual’ objects (circled).
Above. Left: Extended focus of a pressure mounted resistance artery. Right: Segmented volume of the same data. The volume contains objects (cell nuclei) which define the number (i.e. one object/nucleus per cell) and type of cell (i.e. determined by shape, position and orientation). The segmentation was performed using a specialised iterative multi-level thresholding and segmentation routine developed by Dr. Daisheng Luo.

 

Automated 3D-analysis.

The quantitative advantage of imaging modalities can only be fully realised once suitable analysis software is available. Unfortunately, the current state-of-the-art in rendering and segmentation is not equipped to cope with the variability of many biological tissues. Thresholding and segmentation routines are at the heart of many image based measurements. We define 'thresholding' as being the process of selecting intensity ranges; 'segmentation' is defined as the process of extracting an object from a volume. Nuclei stained blood vessels present a particular challenge to the thresholding and segmentation routines. Essentially, the vascular wall can be treated as a 3D volume containing several objects (in this case nuclei) which have different sizes, shapes, orientations and intensities. The challenge is to accurately segment each nucleus from the volume with the minimum input from the user. Several segmentation methods have been suggested. In biomedical image processing, Ong et al. (1996) gives a review of four categories for the segmentation of tissue section images: thresholding, region growing, edge detection, and pattern matching. Most methods deal only with 2-D images and although some can be extended to 3D, it is far more complicated. Specific routines for confocal derived data are particularly difficult to find. The need for such routines has prompted us to develop our own methods designed to handle volumetric data from vascular segments.