Three-dimensional structural analysis from biological confocal images

Laser confocal imaging gives the capability of obtaining 3-D images of biological and other microscopic structures using optical slicing. This eliminates one of the problems of classical scene understanding where a 3-D to 2-D mapping and subsequent loss of information has taken place. In this paper we discuss the implementation of a multidimensional image analysis system used to extract and analyze biological structures in confocal images. The image processing and structure analysis techniques are implemented as a set of knowledge sources or tools that operate on the input image to successively refine and segment the image. These knowledge sources in the system consist of a combination of simple edge/surface detectors and multilevel thresholding to generate the initial segmentation. This segmentation is further refined by a set of more complex knowledge sources such as 3-D morphological operators, a hierarchical refinement scheme, specialized feature detectors, active contour models, and by the use of external knowledge. The analysis tools operate on the segmented image to obtain morphometrical parameters such as surface area, volume, position, and orientation of structures of interest and to generate overall statistical properties of these parameters. A visualization subsystem lets the user manipulate and selectively display the results of segmentation.