A new image segmentation algorithm with applications in confocal microscopy analysis

Gene regulatory networks (GRNs) represent the molecular interactions that govern the behavior of cells in tissues during development. The building and analysis of GRNs require quantitative information on gene expression from tissues. Laser Scanning Confocal Microscopy (LSCM) is commonly used to obtain such information, where immunofluorescence signal can be used as a correlate of gene expression or protein levels. However, a critical step for the extraction of this information is the segmentation of LSCM digital images. Popular segmentation algorithms are frequently based on watershed methods. Here we present an algorithm for the 3D segmentation of nuclei from LSCM (x,y,z) image stacks based on regional merging and graph contractions. This algorithm outperforms watershed methods, especially when the density of images along the z-axis is low and there is a high nuclear signal crowding. In addition, it reduces the parameterization since no filter is needed in order to prevent signal noise side effects (e.g. oversegmentation). Based on this algorithm, we developed an application (iFLIC, immunoFLuorescence Imaging Cytometry tool) for the Java Virtual Machine (JVM). The application supports basic operations for reading, writing and filtering 8-bit depth multicolor TIFF image formats, including indexed file directories (IFD), which are provided by the Java Advanced Imaging (JAI) library. It also provides with basic 3D-rendering and ROI specification that make extensive use of the Java3D library. iFLIC is also a plugin based application powered by the Java Plugin Platform (JPF), so each specific operation is declared as a unique command associated to one plugin and linked to a common interface. Results from segmentation can be exported both as TIFF images and a descriptive file format (iFLIC format)