Some recent works on intercellular communication pointed out an impaired trafficking of Cx43 proteins in early carcinogenesis. In collaboration with biologists, we propose an automatic system for the analysis of spatial protein configurations within cells at early tumor stages. This system is an essential step towards the future development of a computer-aided diagnosis tool and the statistical validation of biological hypotheses about Cx43 expressions and configurations during tumorogenesis. The proposed system contains two dependent part: a segmentation part in which the cell structures of interest are automatically located on images and a characterization part in which some spatial features are computed for the classification of cells. Using immunofluorescent images of cells, the nucleus, cytoplasm and proteins structures within the cell are extracted. Then, some spatial features are computed to characterize spatial configurations of the proteins with regard to the nucleus and cytoplasm areas in the image. Last, the 3D cell images are classified into pathogenic or viable classes. The system has been quantitatively evaluated over 60 cell images acquired by a deconvolution high-resolution microscope and whose ground truth has been manually given by a biologist expert. As a perspective, a 3D spatial reasoning and visualization module is currently under development.