Active Contours Driven by Cellular Neural Networks for Image Segmentation in Biomedical Applications

This paper proposes a novel approach for image segmentation in the context of biomedical applications. The medical images considered for this analysis are both microarray images and images recorded from microfluidics devices. In case of microarray images, microarray spots represented as circular shapes are localised and used further on for the estimation of gene expression levels, based on the average pixel intensities. Considering the microfluidic devices images, the features of cells or clusters of cells are determined by using the proposed image segmentation approach. The novelty of this approach lies in the fact that this is a segmentation procedure which uses an edge-based active contour model (ACM) driven by cellular neural networks (CNNs). Thus, a predefined curve is evolved towards the edges of the image objects (i.e., circular microarray spots and irregular shapes representing clusters of cells). In the curve evolution process, the classic representation of image edges by using the gradient vector is replaced by an edge-based feature template determined by the CNN for each image object. The benefits of the proposed segmentation method are illustrated for both image processing applications, by using specific quality measures for the characterization of object features within the image under analysis.

Automated summary

This paper proposes a novel approach for image segmentation in the context of biomedical applications, using an edge-based active contour model driven by cellular neural networks. This method is used to determine the features of cells or clusters of cells in microarray images and images from microfluidic devices. By replacing the classic representation of image edges with an edge-based feature template, this approach demonstrates benefits for object feature characterization in image processing applications.