Knowledge-driven image analysis of cell structures

Digital Imaging Light Microscopy technology has made major advancements over the past years. Changes have included significant improvements in sensitivity and image quality, but also the advent of innovative methods that have greatly increased the versatility of instruments. Methods to yield contrast in the specimen have made leaps in specificity and signal-to-noise ratio. The combination of these advanced technologies now permits the acquisition of 2 and 3 dimensional image data, of up to five independently labeled parameters, in fixed specimens, or - in time lapse series - of living cells. Given the vast amount of data that can be generated with these instruments, there is a growing need for image analysis software that can yield quantitative measurements from these complex images. The authors are developing a knowledge driven image analysis system that is capable of handling the described image data. The proposed architecture tries to optimize robustness by taking advantage of knowledge that is available both from the expert user as well as from the multiple image sources. It therefore offers significant potential for improvements in performance over linear and mathematical morphology based systems. The authors applied this system to trace actin-based fiber bundles using knowledge constraints, and analyzed their distribution, shape, sizes and orientation in fixed cells by immunofluorescence with good results.